Oracle Redo latch相关测试
测试持有redo copy和allocation latch后的场景：
—-session 1
SYS@PROD1> col name format a30
SYS@PROD1> select addr,latch#,name,level# from v$latch where name like '%redo%';
ADDR LATCH# NAME LEVEL#
-------- ---------- ------------------------------ ----------
2000DF34 146 redo on-disk SCN 8
2000DF98 147 ping redo on-disk SCN 8
2001252C 185 redo writing 1
20012594 186 redo copy 4
200125FC 187 redo allocation 5
20012664 188 real redo SCN 8
20012CD0 189 readredo stats and histogram 8
200149F0 209 readable standby metadata redo 0
cache
200264C0 515 KFR redo allocation latch 6
9 rows selected.
SYS@PROD1> select to_number('20012594','XXXXXXXXXX') from dual;
TO_NUMBER('20012594','XXXXXXXXXX')
----------------------------------
536946068
SYS@PROD1>
SYS@PROD1>
SYS@PROD1>
SYS@PROD1> oradebug setmypid
Statement processed.
SYS@PROD1> oradebug call kslgetl 536946068 1
Function returned 1
SYS@PROD1> oradebug call kslgetl 536946172 1
Function returned 1
SYS@PROD1>
—-session 2
SYS@PROD1> delete from test01 where rownum < 10;
9 rows deleted.
SYS@PROD1> commit;
Commit complete.
SYS@PROD1> select to_number('200125FC','XXXXXXXXXX') from dual;
TO_NUMBER('200125FC','XXXXXXXXXX')
----------------------------------
536946172
SYS@PROD1> delete from test01 where rownum < 10;
9 rows deleted.
SYS@PROD1> commit;
Commit complete.
SYS@PROD1>
SYS@PROD1> select count(*) from v$latch; ---------------------> hang住
select count(*) from v$latch
*
ERROR at line 1:
ORA-12152: TNS:unable to send break message
SYS@PROD1>
alert报警日志：
Tue Dec 13 15:21:48 2016
Exception [type: SIGSEGV, Address not mapped to object] [ADDR:0xC7] [PC:0xDFDCCB7, kcrfw_copy_latch_cleanup()+235] [flags: 0x0, count: 1]
Errors in file /u01/app/oracle/diag/rdbms/prod1/PROD1/trace/PROD1_ora_14934.trc (incident=26591):
ORA-07445: exception encountered: core dump [kcrfw_copy_latch_cleanup()+235] [SIGSEGV] [ADDR:0xC7] [PC:0xDFDCCB7] [Address not mapped to object] []
ORA-07443: function kslfree not found
Incident details in: /u01/app/oracle/diag/rdbms/prod1/PROD1/incident/incdir_26591/PROD1_ora_14934_i26591.trc
Use ADRCI or Support Workbench to package the incident.
See Note 411.1 at My Oracle Support for error and packaging details.
Tue Dec 13 15:21:51 2016
Dumping diagnostic data in directory=[cdmp_20161213152151], requested by (instance=1, osid=14934), summary=[incident=26591].
Tue Dec 13 15:21:52 2016
1）oracle里面是不允许同时去持有父latch和子latch的，如果你时去持有子latch，我们会发现实例会down掉。
2）redo allocation latch 这个latch的获取是在server process将redo entry copy到redo buffer（shared private strands）中之前，需要获得该latch，以便于分配空间，用于存放redo entry.
redo allocation latch 相当多
SYS@PROD1> select addr,name,latch#,level#,child#,hash from v$latch_children where latch# in (select latch# from v$latch where name like '%redo%');
ADDR NAME LATCH# LEVEL# CHILD# HASH
-------- -------------------- ---------- ---------- ---------- ----------
36D1EF44 redo copy 186 4 4 4092072627
36D1EEC0 redo copy 186 4 3 4092072627
36D1EE3C redo copy 186 4 2 4092072627
36D1EDB8 redo copy 186 4 1 4092072627
36D218D0 redo allocation 187 5 29 999804931
36D2186C redo allocation 187 5 28 999804931
36D21808 redo allocation 187 5 27 999804931
36D217A4 redo allocation 187 5 26 999804931
36D21740 redo allocation 187 5 25 999804931
36D216DC redo allocation 187 5 24 999804931
36D21678 redo allocation 187 5 23 999804931
36D21614 redo allocation 187 5 22 999804931
36D215B0 redo allocation 187 5 21 999804931
36D2154C redo allocation 187 5 20 999804931
36D214E8 redo allocation 187 5 19 999804931
36D21484 redo allocation 187 5 18 999804931
36D21420 redo allocation 187 5 17 999804931
36D213BC redo allocation 187 5 16 999804931
36D21358 redo allocation 187 5 15 999804931
36D212F4 redo allocation 187 5 14 999804931
36D21290 redo allocation 187 5 13 999804931
36D2122C redo allocation 187 5 12 999804931
36D211C8 redo allocation 187 5 11 999804931
36D21164 redo allocation 187 5 10 999804931
36D21100 redo allocation 187 5 9 999804931
36D2109C redo allocation 187 5 8 999804931
36D21038 redo allocation 187 5 7 999804931
36D20FD4 redo allocation 187 5 6 999804931
36D20F70 redo allocation 187 5 5 999804931
36D20F0C redo allocation 187 5 4 999804931
36D20EA8 redo allocation 187 5 3 999804931
36D20E44 redo allocation 187 5 2 999804931
36D20DE0 redo allocation 187 5 1 999804931
33 rows selected.
SYS@PROD1>
3） redo writing latch
如果该latch会持有，那么必然导致lgwr无法操作，因为当事务commit后，会触发lgwr进程将redo buffer中的脏数据写入到redo logfile中，而lgwr在写之前，必须先获得redo writing latch。
—- session 1
SYS@PROD1> col name format a30
SYS@PROD1> select addr,latch#,name,level# from v$latch where name like '%redo%';
ADDR LATCH# NAME LEVEL#
-------- ---------- ------------------------------ ----------
2000DF34 146 redo on-disk SCN 8
2000DF98 147 ping redo on-disk SCN 8
2001252C 185 redo writing 1
20012594 186 redo copy 4
200125FC 187 redo allocation 5
20012664 188 real redo SCN 8
20012CD0 189 readredo stats and histogram 8
200149F0 209 readable standby metadata redo 0
cache
200264C0 515 KFR redo allocation latch 6
9 rows selected.
SYS@PROD1> oradebug setmypid
Statement processed.
SYS@PROD1> select to_number('2001252C','XXXXXXXXXXXX') from dual;
TO_NUMBER('2001252C','XXXXXXXXXXXX')
------------------------------------
536945964
SYS@PROD1> oradebug call kslgetl 536945964 1
Function returned 1
SYS@PROD1>
—- session 2
SYS@PROD1> delete from test01 where rownum < 2;
1 row deleted.
SYS@PROD1> commit;
----> commit 之后 hang住在这里
—- session 1
SYS@PROD1> oradebug call kslfre 536945964
Function returned 0
SYS@PROD1>
—- session 2
SYS@PROD1> commit;
Commit complete.
SYS@PROD1>
释放掉之后redo write latch之后就可以commit了