zoukankan      html  css  js  c++  java
  • 10046确认多块读无法跨区

    SQL查询记录,记录是如何从磁盘中读取的呢? 有几种访问方式?

    本次测试探讨数据文件离散读无法跨区,因此单块读(顺序读)存在特殊场景,一个表的一个块在某个区中,Oracle查询记录及时全表扫描,也会产生单块读现象。

    文档结构如下:

    1.解释说明oracle数据文件从磁盘读取有几种方式

    2.创建测试对象,并进行多块读测试解释说明

    3.创建测试环境,一个表已分配多个区,存在一个区只使用一个块,次数对测试表全表扫描,验证是否存在单块读现象。

    一、解释说明oracle数据文件从磁盘读取有几种方式

        EVENT#   EVENT_ID NAME                                PARAMETER1    PARAMETER2 PARAMETER3 WAIT_CLASS
    ---------- ---------- ----------------------------------- ------------- ---------- ---------- ----------
             9 2171045634 remote db file read                 clientid      count      intr       Network
           146 2652584166 db file sequential read             file#         block#     blocks     User I/O
           147  506183215 db file scattered read              file#         block#     blocks     User I/O
           151  834992820 db file parallel read               files         blocks     requests   User I/O
           197 3926164927 direct path read                    file number   first dba  block cnt  User I/O
           198  861319509 direct path read temp               file number   first dba  block cnt  User I/O
           269 3999678875 securefile direct-read completion                                       User I/O
    
    7 rows selected.

    三种:方式依次为 多块读,单块读,直接路径读

    db file scattered read

    db file sequential read

    direct path read   

    多块读:一次IO读取多个块,读取无法跨区,常见与全表扫描

    单块读:一次IO读取单个块,常见索引rowid回表提取记录,非常见全表扫描有个块跨区落单了

    直接路径读:读取的数据块,不放入buffer cache中,直接提取到会话PGA中,常见与LOB字段读取


      

    二、多块读测试

    1)创建测试数据
    SQL> create table scott.e(id char(2000),name char(2000),
    deptno char(2000),salary char(2000)); Table created. 2)插入测试数据 declare v_id int; begin for v_id in 1 .. 7 loop insert into scott.e values(v_id,v_id||'name',
    v_id||'deptno',v_id||'salary'); end loop; commit; end; / 3)查询数据存储记录
    SQL> select tablespace_name,file_id,extent_id,block_id,
        bytes/1024 bytes_k,blocks from dba_extents
        where owner='SCOTT' and segment_name='E'
    
    TABLESPACE_NAM    FILE_ID  EXTENT_ID   BLOCK_ID    BYTES_K     BLOCKS
    -------------- ---------- ---------- ---------- ---------- ----------
    USERS                   4          0       1768         64          8
    USERS                   4          1       1776         64          8
    USERS                   4          2       1784         64          8


    4)查询rowid
    SQL> select rowid,substr(id,1,2) from scott.e; ROWID SUBSTR(ID,1,2) ------------------ ---------------- AAAS5RAAEAAAAbsAAA 2 AAAS5RAAEAAAAbvAAA 1 AAAS5RAAEAAAAbwAAA 3 AAAS5RAAEAAAAbxAAA 4 AAAS5RAAEAAAAb0AAA 6 AAAS5RAAEAAAAb2AAA 5 AAAS5RAAEAAAAb/AAA 7 7 rows selected.
    rowid解释说明,从左到有,可以发现表7行数据占用7个块

    数据对象编号(6位显示)
    相关数据文件编号(3位显示)
    数据块编号(6位显示)
    数据块中行编号(3位显示)

      

    5)根据rowid查询块编号
    select rowid, dbms_rowid.rowid_object(rowid) "objct", dbms_rowid.rowid_relative_fno(rowid) "file", dbms_rowid.rowid_block_number(rowid) "block", dbms_rowid.rowid_row_number(rowid) "row" from scott.e; ROWID objct file block row ------------------ --------- ---------- ---------- ---------- AAAS5RAAEAAAAbsAAA 77393 4 1772 0 AAAS5RAAEAAAAbvAAA 77393 4 1775 0 AAAS5RAAEAAAAbwAAA 77393 4 1776 0 AAAS5RAAEAAAAbxAAA 77393 4 1777 0 AAAS5RAAEAAAAb0AAA 77393 4 1780 0 AAAS5RAAEAAAAb2AAA 77393 4 1782 0 AAAS5RAAEAAAAb/AAA 77393 4 1791 0 7 rows selected.

      6)数据文件离散读测试

    SQL> alter system flush buffer_cache;


    SQL> alter session set events '10046 trace name context forever,level 12';

    Session altered.

    SQL> select count(*) from scott.e;

    COUNT(*)
    ----------
    7

    SQL> alter session set events '10046 trace name context off';

    Session altered.

    jx1:/u01/app/oracle/diag/rdbms/jx/jx1/trace$ cat jx1_ora_30782.trc |grep scatt
    WAIT #140130911224448: nam='db file scattered read' ela= 713 file#=4 block#=1771 blocks=5 obj#=77393 tim=1545603274223872
    WAIT #140130911224448: nam='db file scattered read' ela= 735 file#=4 block#=1776 blocks=8 obj#=77393 tim=1545603274224854
    WAIT #140130911224448: nam='db file scattered read' ela= 1281 file#=4 block#=1785 blocks=7 obj#=77393 tim=1545603274226337 

    可以发现执行一次全表扫描,数据文件多块读,读取了三次,每次返回不同的块数量

      

    三、创建测试环境,一个表已分配多个区,存在一个区只使用一个块,次数对测试表全表扫描,验证是否存在单块读现象

    1)环境构建,使用笨拙的办法,不断insert,查询extent分配

    SQL> insert into scott.e values(8,8,8,8);

    SQL> insert into scott.e values(9,9,9,9);

    SQL> insert into scott.e values(10,10,10,10);

    SQL> insert into scott.e values(11,11,11,11);

    SQL> insert into scott.e values(12,12,12,12);

    SQL> r
    1 select tablespace_name,file_id,extent_id,block_id,
    2 bytes/1024 bytes_k,blocks from dba_extents
    3 where owner='SCOTT' and segment_name='E'
    4*

    TABLESPACE_NAM FILE_ID EXTENT_ID BLOCK_ID BYTES_K BLOCKS
    -------------- ---------- ---------- ---------- ---------- ----------
    USERS 4 0 1768 64 8
    USERS 4 1 1776 64 8
    USERS 4 2 1784 64 8
    USERS 4 3 3200 64 8  --终于新扩建一个区

    SQL> select rowid from scott.e where id like '%12%';

    ROWID
    ------------------
    AAAS5RAAEAAAAyGAAA

    select rowid,
     dbms_rowid.rowid_object(rowid) "objct",
     dbms_rowid.rowid_relative_fno(rowid) "file",
     dbms_rowid.rowid_block_number(rowid) "block",
     dbms_rowid.rowid_row_number(rowid) "row"
    from scott.e;
    ROWID                   objct       file      block        row
    ------------------ ---------- ---------- ---------- ----------
    AAAS5RAAEAAAAbsAAA      77393          4       1772          0
    AAAS5RAAEAAAAbvAAA      77393          4       1775          0
    AAAS5RAAEAAAAbwAAA      77393          4       1776          0
    AAAS5RAAEAAAAbxAAA      77393          4       1777          0
    AAAS5RAAEAAAAb0AAA      77393          4       1780          0
    AAAS5RAAEAAAAb2AAA      77393          4       1782          0
    AAAS5RAAEAAAAb7AAA      77393          4       1787          0
    AAAS5RAAEAAAAb9AAA      77393          4       1789          0
    AAAS5RAAEAAAAb/AAA      77393          4       1791          0
    AAAS5RAAEAAAAb/AAB      77393          4       1791          1
    AAAS5RAAEAAAAb/AAC      77393          4       1791          2
    AAAS5RAAEAAAAyGAAA      77393          4       3206          0     ---
    
    12 rows selected.
    

      

    2)10046跟踪

    SQL> alter system flush buffer_cache;

    SQL> alter session set events '10046 trace name context forever,level 12';

    Session altered.

    SQL> select count(*) from scott.e;

    COUNT(*)
    ----------
    12

    SQL> alter session set events '10046 trace name context off';

    Session altered.

    jx1:/u01/app/oracle/diag/rdbms/jx/jx1/trace$ cat jx1_ora_55917.trc|grep read
    Redo thread mounted by this instance: 1
    WAIT #139784104305960: nam='db file sequential read' ela= 15423 file#=4 block#=1770 blocks=1 obj#=77393 tim=1545605262033100 (OK)
    WAIT #139784104305960: nam='db file scattered read' ela= 650 file#=4 block#=1771 blocks=5 obj#=77393 tim=1545605262034068
    WAIT #139784104305960: nam='db file scattered read' ela= 708 file#=4 block#=1776 blocks=8 obj#=77393 tim=1545605262035050
    WAIT #139784104305960: nam='db file scattered read' ela= 678 file#=4 block#=1785 blocks=7 obj#=77393 tim=1545605262036010
    WAIT #139784104305960: nam='db file scattered read' ela= 12389 file#=4 block#=3200 blocks=8 obj#=77393 tim=1545605262048744

    原谅我容易满足,得到单块读现象后,我现在停止了脚步,如有兴趣的朋友,可以继续研究,这些块都是什么对象,为什么这读取,是否存在什么规则。

      

  • 相关阅读:
    如何用UE(UltraEdit)删除重复行?--转
    spring源码分析之<context:component-scan/>vs<annotation-config/>
    annotation-config vs component-scan – Spring Core--转
    spring源码分析之spring注解@Aspect是如何工作的?
    Spring之LoadTimeWeaver——一个需求引发的思考---转
    Linux上的free命令详解
    MVC/MVP/MVVM区别——MVVM就是angular,视图和数据双向绑定
    elasticsearch如何安全重启节点
    ES等待任务——是master节点上的task任务
    1. 批量梯度下降法BGD 2. 随机梯度下降法SGD 3. 小批量梯度下降法MBGD
  • 原文地址:https://www.cnblogs.com/lvcha001/p/10203053.html
Copyright © 2011-2022 走看看