继续验证,使用solaris中的dtrace
session1:编写dtrace脚本
more test.d #!/usr/sbin/dtrace -s -n dtrace:::BEGIN { i=1; } pid$1:::entry { printf("i=%d PID::entry:==%s:%s:%s:%s %x %x %x %x %x %x",i, probeprov, probemod, probefunc, probename,arg0,arg1,arg2,arg3,arg4,arg5); i=i+1; }
session2:多执行几次rowid查询,使查询走逻辑读
create table a as select * from scott.emp; create index ind_a on a(empno); select c.sid,spid,pid,a.SERIAL# from (select sid from v$mystat where rownum<=1) c,v$session a,v$process b where c.sid=a.sid and a.paddr=b.addr; SID SPID PID SERIAL# ---------- ------------ ---------- ---------- 159 1154 15 3 select * from a where rowid='AAAMjsAABAAAOs6AAA';
session1:运行dtrace脚本,并输出到日志
./test.d 1154 > logic_read1.log
session2:再次执行一次,此时为逻辑读
select * from a where rowid='AAAMjsAABAAAOs6AAA';
session1:查询文件,文件内容为语句执行期间做的跟踪
cat logic_read1.log|wc -l 1393
session2:
select dbms_rowid.ROWID_RELATIVE_FNO('AAAMjsAABAAAOs6AAA'),dbms_rowid.rowid_block_number('AAAMjsAABAAAOs6AAA') from dual; 1 60218 select file#,dbablk,tch,lower(HLADDR) from x$bh where file#=1 and dbablk=60218; FILE# DBABLK TCH LOWER(HL ---------- ---------- ---------- -------- 1 60218 8 5f0f596c
1号文件60218号块,是受地址为5f0f596c的Latch保护
session1:
cat logic_read1.log|grep 5f0f596c 1 113271 sskgslcas:entry i=637 PID::entry:==pid1154:oracle:sskgslcas:entry 5f0f596c 0 2000000f fdc1a128 fdc1a0d0 fdc1a128 1 113274 sskgsldecr:entry i=646 PID::entry:==pid1154:oracle:sskgsldecr:entry 5f0f596c 2000000f fdc1a128 fdc1a0d0 fdc1a128 804534c 1 113271 sskgslcas:entry i=687 PID::entry:==pid1154:oracle:sskgslcas:entry 5f0f596c 0 2000000f 1 fdc1a0c0 5854e074 1 59436 kcbzar:entry i=692 PID::entry:==pid1154:oracle:kcbzar:entry 5f181bcc 5f0f596c 108000 8045268 1 fdc1a0c0 1 103456 slmxnoop:entry i=693 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 103456 slmxnoop:entry i=694 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 103456 slmxnoop:entry i=695 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 103456 slmxnoop:entry i=696 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 103456 slmxnoop:entry i=697 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 103456 slmxnoop:entry i=699 PID::entry:==pid1154:oracle:slmxnoop:entry 587ed578 fdc1a130 5f0f596c 8045238 a9ef965 c4a88d0 1 113274 sskgsldecr:entry i=701 PID::entry:==pid1154:oracle:sskgsldecr:entry 5f0f596c 2000000f 1 fdc1a0c0 5854e074 8045410 1 54264 kssrmf:entry i=703 PID::entry:==pid1154:oracle:kssrmf:entry 5f181ba8 5f0bcc2c 587ed578 2000000f 5f0f596c 5f181bcc
和这个地址(5f0f596c)相关的有这十几行。在这里,有一点编程习惯再说一下,要申请某一个地址处的Latch,这个Latch的地址,是这个函数的最重要的参数,
因此,Oracle会把它排在第一位,也就是说,以上这十几行中,第一个参数不是8ea1d750的,基本可以排除掉了。
所以,我们只剩这些行需要关注:
1 113271 sskgslcas:entry i=637 PID::entry:==pid1154:oracle:sskgslcas:entry 5f0f596c 0 2000000f fdc1a128 fdc1a0d0 fdc1a128 1 113274 sskgsldecr:entry i=646 PID::entry:==pid1154:oracle:sskgsldecr:entry 5f0f596c 2000000f fdc1a128 fdc1a0d0 fdc1a128 804534c 1 113271 sskgslcas:entry i=687 PID::entry:==pid1154:oracle:sskgslcas:entry 5f0f596c 0 2000000f 1 fdc1a0c0 5854e074 1 113274 sskgsldecr:entry i=701 PID::entry:==pid1154:oracle:sskgsldecr:entry 5f0f596c 2000000f 1 fdc1a0c0 5854e074 8045410
这四行,两个函数调用,sskgslcas、sskgsldecr,
第一个参数都是Latch的地址:5f0f596c 我相信这不是巧合,它们肯定是申请、释放Latch的函数。
i=637这行,Oracle调用sskgslcas持有Latch,在i=646这行,调用sskgsldecr释放,
接下来在i=687又一次调用sskgslcas持有Latch,在i=701处调用sskgsldecr释放。
一次逻辑读对应两次Latch调用。
结果是这样吗,让我们继续验证,Oracle的Oradebug可以调用某个Oracle自身的函数,就有它来验证吧:
session2:
SQL> oradebug setmypid Statement processed. SQL> oradebug call sskgslcas 0x5f0f596c 0 0x2000000f 0xfdc1a128 Function returned 1 SQL>select * from a where rowid='AAAMjsAABAAAOs6AAA'; Hang住了
session3:查看等等事件,并释放latch
SQL> select sid,event,p1raw,p2 from v$session where sid=159; SID EVENT ---------- ---------------------------------------------------------------- P1RAW P2 -------- ---------- 159 latch: cache buffers chains 5F0F596C 122 SQL> oradebug setospid 1154 Oracle pid: 15, Unix process pid: 1154, image: oracle@sol (TNS V1-V3) SQL> oradebug call sskgsldecr 0x5f0f596c 0x2000000f Function returned 2000000F
在session2中查看,此时结果出来了。说明被Hang的会话,已经可以顺利执行下去了。说明Latch已经被释放。
结论:一次逻辑读,两次cbc latch,cbc latch的申请函数为sskgslcas,释放函数为sskgsldecr
逻辑读时的cbc latch(一):http://qhd2004.blog.51cto.com/629417/1576554
逻辑读时的cbc latch(三): http://qhd2004.blog.51cto.com/629417/1577532
参考:http://blog.itpub.net/321157/viewspace-730936/