深入解析bufferbusywaits

在写一个培训ppt的时候，为了深入理解buffe busy waits这个等待事件，做了一个仔细的测试，对大家也有帮助，经过测试，发现我个人以前的认识都有一点问题。大家一起探讨！ 1. 创建测试表 http://www.killdb.comconn roger/rogerConnected.www.killdb.comcreate tabl

在写一个培训ppt的时候，为了深入理解buffe busy waits这个等待事件，做了一个仔细的测试，对大家也有帮助，经过测试，发现我个人以前的认识都有一点问题。大家一起探讨！

1. 创建测试表

www.killdb.com>conn roger/rogerConnected.www.killdb.com>create table t_buffer_busy_waits as select * from dba_objects where rownum  create index t on t_buffer_busy_waits(object_id);Index created.www.killdb.com>select dbms_rowid.rowid_relative_fno(rowid) fn#,2         dbms_rowid.rowid_block_number(rowid) blk#,3         object_id4    from t_buffer_busy_waits5   where object_id = 1006  /FN#       BLK#  OBJECT_ID---------- ---------- ----------5      28909        100www.killdb.com>select dbms_rowid.rowid_relative_fno(rowid) fn#,2         dbms_rowid.rowid_block_number(rowid) blk#,3         object_id4    from t_buffer_busy_waits5   where object_id = 1016  /FN#       BLK#  OBJECT_ID---------- ---------- ----------5      28909        101www.killdb.com>c/101/995*  where object_id = 99www.killdb.com>/FN#       BLK#  OBJECT_ID---------- ---------- ----------5      28909         99www.killdb.com>c/99/1115*  where object_id = 111www.killdb.com>/FN#       BLK#  OBJECT_ID---------- ---------- ----------5      28909        111

注意，我这里就拿这几条在同一个block内的数据，来进行测试模拟 2. 模拟高并发读取 模拟高并发读取的测试过程中，我同时开了3个窗口，进行测试，另外还开了第4个窗口进行检测event，如下：

--session 1www.killdb.com>select sid from v$Mystat where rownum declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      select count(*) into c from t_buffer_busy_waits where object_id = 99;6    end loop;7  end;8  /PL/SQL procedure successfully completed.--session 2www.killdb.com>select sid from v$Mystat where rownum www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      select count(*) into c from t_buffer_busy_waits where object_id = 100;6    end loop;7  end;8  /PL/SQL procedure successfully completed.--session 3www.killdb.com>select sid from v$mystat where rownum declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      select count(*) into c from t_buffer_busy_waits where object_id = 111;6    end loop;7  end;8  /PL/SQL procedure successfully completed.www.killdb.com>

通过第4个会话窗口的检测，我发现，并没有buffer busy waits等待出现，但是出现了大家非常熟悉的latch：cache buffer chains.
监控会话（session 4）：

ww.killdb.com>select event,count(1) from v$session where wait_class#6 group by event;EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1latch: cache buffers chains                                               1www.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1latch: cache buffers chains                                               1

从第一个测试来看，并发读取实际上并不会产生buffer busy waits。
3. 模拟2个并发会话，分别为读与写

--session 1www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      select count(*) into c from t_buffer_busy_waits where object_id = 99;6    end loop;7  end;8  /PL/SQL procedure successfully completed.www.killdb.com>---session 2www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      update  t_buffer_busy_waits set object_id=111 where object_id=111;6    end loop;7  end;8  /declare*ERROR at line 1:ORA-30036: unable to extend segment by 8 in undo tablespace 'UNDOTBS2'ORA-06512: at line 5

大家注意，这里测试的时候必须使用位于同一个数据块内的行，前面我们已经知道object_id=99和111的这2条数据是在
同一个数据块内（实际上，object_id 从99到111都是在同一块内，不用说明了吧？）
ok，下面我们来看下第3个监控会话的信息是什么样的？

---session 3www.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1control file sequential read                                              1log file switch completion                                                1。。。。。EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1latch: cache buffers chains                                               1log file switch completion                                                1www.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1latch: cache buffers chains                                               2。。。。。。www.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1

比较奇怪，居然没有出现应该出现的buffer busy waits等待事件。难道读和写不会出现这个等待吗 ？
我们先保留这个疑问，继续测试写和写的并发是否会产生buffer busy waits。
4. 模拟2个会话并发同时写

---Session 1www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=99 where object_id=99;6    end loop;7  end;8  /PL/SQL procedure successfully completed.--Session 2www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=111 where object_id=111;6    end loop;7  end;8  /PL/SQL procedure successfully completed.

ok，下面我们来看下第3个监控会话的结果：

www.killdb.com>l1* select event,count(1) from v$session where wait_class#6 group by eventwww.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1buffer busy waits                                                         1latch: cache buffers chains                                               1www.killdb.com>/EVENT                                                              COUNT(1)---------------------------------------------------------------- ----------SQL*Net message to client                                                 1buffer busy waits                                                         1latch: cache buffers chains                                               1www.killdb.com>select sid,username,p1,p2,p3,event from v$session where event like 'buffer%';SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------523 ROGER               5      28909          1 buffer busy waits

ok，终于出现了我们希望的结果了，这个buffer busy waits来的不容易。这里我们需要说明一点，从oracle 10.1版本开始，
这里的p3 不再是像之前的数据库版本中那样，用来表示buffer busy waits的reason了，这里表示的是所等待的block的数据块
类型。很明显，这里的p3为1，那么也就是说block type 为1，表示data block，这一点不用多说了吧！
大家看Oracle AWR，我想也经常看到buffer waits，其实也会出现在其他的block 类型上，不仅仅是data block。
只不过数据块的争用更为常见一些。如果你查询文档你会发现，有下面的一些block type类型。

1 data block        7 extent map    13 file header block2 sort block        8 1st level bmb 14 unused3 save undo block   9 2nd level bmb 15+2*x undo header block(x=usn#)4 segment header   10 3rd level bmb 16+2*x undo block(x=usb#)5 save undo header 11 bitmap block6 free list        12 bitmap index block

既然有可能是其他类型的block也会出现，那么按理说读和写是不兼容的，也应该出现，难道我们的测试有问题？
可能是并发不够，这里我再次测试了第2个测试：

Session 1：www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=105 where object_id=105;6    end loop;7  end;8  /session 2：www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=106 where object_id=106;6    end loop;7  end;8  /session 3:www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=107 where object_id=107;6    end loop;7  end;8  /Session 4:www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 500000 loop5      update  t_buffer_busy_waits set object_id=108 where object_id=108;6    end loop;7  end;8  /Session 5:www.killdb.com>declare2    c number;3  begin4    for i in 1 .. 1000000 loop5      select count(*) into c from t_buffer_busy_waits where object_id =109;6    end loop;7  end;8  /

我这里同时模拟了5个会话进行操作，其中4个是update，一个进行select查询，注意，这里重复操作的5条数据，其实都是在同一个数据块中。
下面我们来看下第6个监控会话的结果是什么呢？

session 6:www.killdb.com>select sid,username,p1,p2,p3,event from v$session where event like 'buffer%';SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               2       5657         23 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               2       5657         23 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               2       5657         23 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------<i style="color:transparent">本文来源gaodai$ma#com搞$代*码6网</i>- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               2       5657         23 buffer busy waitsw.killdb.com>select sid,username,p1,p2,p3,event from v$session where event like 'buffer%';SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------534 ROGER               5      28909          1 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               5      28909          1 buffer busy waits523 ROGER               5      29037          1 buffer busy waits524 ROGER               5      29037          1 buffer busy waits537 ROGER               5      29037          1 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------521 ROGER               5      28909          1 buffer busy waits523 ROGER               5      29037          1 buffer busy waits524 ROGER               5      29037          1 buffer busy waits537 ROGER               5      29037          1 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------520 ROGER               5      29037          1 buffer busy waits521 ROGER               5      28909          1 buffer busy waits534 ROGER               5      29037          1 buffer busy waitswww.killdb.com>/SID USERNAME           P1         P2         P3 EVENT---------- ---------- ---------- ---------- ---------- ----------------------------------------520 ROGER               5      29037          1 buffer busy waits521 ROGER               5      28909          1 buffer busy waits534 ROGER               5      29037          1 buffer busy waits

我们可以看到，不仅仅是数据块，这里undo 也出现的buffer busy waits，很明显，这里的file 2是undo datafile。
因此，下面我们来个简单的总结：
1) Buffer busy waits等待的本质是因为写的缘故出现争用.
2) Oracle里面写写是不兼容的，写和读也是不兼容的，有可能出现buffer busy waits等待.
3) 实际上读写并发容易产生回滚段的争用，等待事件也是buffer busy waits.

喜欢 (0)赏

[搞代码]

分享 (0)