Why doesn't AlTER SYSTEM SET EVENTS set the events or tracing immediately?

Why doesn't AlTER SYSTEM SET EVENTS set the events or tracing immediately?

  今天看到Tanel的blog，说明了此问题：
  我们通过alter system 设置events和parameter是不同的，
对于设置parameter使用alter system的影响如下：
1）对当前的session起作用
2）对新用户登录数据库的用户起作用
3）对已经登录到数据库的其他session起作用
对于使用alter system 设置events影响如下：
1）对当前session起作用
2）对新登录到系统中的用户起作用，但不对其他的session起作用。
1、对于alter system set parameters验证如下：
eg：
目前系统只有三个session登录：

SQL> select saddr,sid,serial#,username from v$session where username is not null;SADDR                   SID    SERIAL# USERNAME---------------- ---------- ---------- ------------------------------000000007CFED5D8         33        103 SYS000000007CFDE178         38        129 SCOTT000000007CFD4ED8         41         65 RHYSSQL> 

首先修改log_checkpoints_to_alert参数为true，当前用户为sys；

SQL> show parameter log_checkpointNAME                                 TYPE        VALUE------------------------------------ ----------- ------------------------------log_checkpoint_interval              integer     0log_checkpoint_timeout               integer     1800log_checkpoints_to_alert             boolean     FALSESQL> alter system set log_checkpoints_to_alert=true;System altered.SQL> SQL> show userUSER is "SYS"SQL> 

在rhys账户查看该parameter值：

SQL> show userUSER is "RHYS"SQL> select * from v$mystat where rownum<3;       SID STATISTIC#      VALUE---------- ---------- ----------        41          0          0        41          1         12SQL> show parameter log_checkpointsNAME                                 TYPE        VALUE------------------------------------ ----------- ------------------------------log_checkpoints_to_alert             boolean     TRUESQL> 

 

在scott账户查看该parameter值：

SQL> show userUSER is "SCOTT"SQL> select * from v$mystat where rownum<3;       SID STATISTIC#      VALUE---------- ---------- ----------        38          0          0        38          1         12SQL> show parameter log_checkpointsNAME                                 TYPE        VALUE------------------------------------ ----------- ------------------------------log_checkpoints_to_alert             boolean     TRUESQL> 

新登录一个用户Amy；

SQL> create user amy identified by rhys default tablespace rhys;User created.SQL> grant dba to amy;Grant succeeded.SQL> select saddr,sid,serial#,username from v$session where username is not null;SADDR                   SID    SERIAL# USERNAME---------------- ---------- ---------- ------------------------------000000007C8836E0          1         83 AMY000000007CFED5D8         33        103 SYS000000007CFDE178         38        137 SCOTT000000007CFD4ED8         41         65 RHYSSQL> SQL> show user USER is "AMY"SQL> select * from v$mystat where rownum<3;       SID STATISTIC#      VALUE---------- ---------- ----------         1          0          0         1          1         12SQL> show parameter log_checkpoints_to_alertNAME                                 TYPE        VALUE------------------------------------ ----------- ------------------------------log_checkpoints_to_alert             boolean     TRUESQL> 

 

2、对于对于alter system set events验证如下：

SQL> select saddr,sid,serial#,username from v$session where username is not null;SADDR                   SID    SERIAL# USERNAME---------------- ---------- ---------- ------------------------------000000007CFED5D8         33        103 SYS000000007CFDE178         38        137 SCOTT000000007CFD4ED8         41         65 RHYSSQL> show userUSER is "SYS"SQL> alter system set events '10046 trace name context forever,level 12';System altered.SQL> SQL> select saddr,sid,s.serial#,s.username,spid from v$session s,v$process p where s.paddr=p.addr and  s.username is not null;SADDR                   SID    SERIAL# USERNAME                       SPID---------------- ---------- ---------- ------------------------------ ------------------------000000007CFDE178         38        137 SCOTT                          2176000000007CFED5D8         33        103 SYS                            1981000000007CFD4ED8         41         65 RHYS                           2092SQL> SQL> !    [oracle@oracle-one ~]$ cd /opt/app/oracle/diag/rdbms/rhys/RHYS/trace/[oracle@oracle-one trace]$ ls -l *1981*-rw-r-----. 1 oracle oinstall 23897 Oct  9 09:58 RHYS_ora_1981.trc-rw-r-----. 1 oracle oinstall   390 Oct  9 09:58 RHYS_ora_1981.trm[oracle@oracle-one trace]$ 

 

可以看到sys用户已经存在trace文件了。
然后在rhys用户下查看一下信息：

SQL> select * from v$mystat where rownum<3;       SID STATISTIC#      VALUE---------- ---------- ----------        41          0          0        41          1         19SQL> SQL> select table_name from user_tables;TABLE_NAME------------------------------DEPTBONUSAMYABCTSYS_TEMP_FBTSALGRADEEMP10 rows selected.SQL> ![oracle@oracle-one ~]$ cd /opt/app/oracle/diag/rdbms/rhys/RHYS/trace/[oracle@oracle-one trace]$ ls -ltr *2092*-rw-r-----. 1 oracle oinstall   1211 Oct  9 10:02 RHYS_ora_2092.trm-rw-r-----. 1 oracle oinstall 164508 Oct  9 10:02 RHYS_ora_2092.trc[oracle@oracle-one trace]$ 

同样也存在trace文件了。
那么对于新增的用户呢？我们使用Amy账户登录。

SQL> show userUSER is "AMY"SQL>  select saddr,sid,s.serial#,s.username,spid from v$session s,v$process p where s.paddr=p.addr and  s.username is not null;SADDR                   SID    SERIAL# USERNAME                       SPID---------------- ---------- ---------- ------------------------------ ------------------------000000007CFDE178         38        137 SCOTT                          2176000000007CFED5D8         33        103 SYS                            1981000000007CFD4ED8         41         65 RHYS                           2092000000007CFC2998         47        137 AMY                            5500SQL> select * from v$mystat where rownum<3;       SID STATISTIC#      VALUE---------- ---------- ----------        47          0          0        47          1         18SQL> ![oracle@oracle-one ~]$ cd /opt/app/oracle/diag/rdbms/rhys/RHYS/trace/[oracle@oracle-one trace]$ ls -l *5500*-rw-r-----. 1 oracle oinstall 184626 Oct  9 10:25 RHYS_ora_5500.trc-rw-r-----. 1 oracle oinstall   1095 Oct  9 10:25 RHYS_ora_5500.trm[oracle@oracle-one trace]$ 

发现也是可以跟踪到trace。

可知在11g中，alter system set events也是对已经登录的会话生效的。但是在10G确不是。
note：
如果在不知情的情况下设置了跟踪事件，想想那是多么可怕的事情。可能导致磁盘占满，系统宕机等事故。
如下是转自tanel的一句话：
This means that the existing, already logged in sessions, will not pick up any of the events set via ALTER SYSTEM!
 
This hopefully explains why sometimes the debug events don’t seem to work. But more importantly, this also means that when you disable an event (by setting it to “OFF” or to level 0) with ALTER SYSTEM, it does not affect the existing sessions who have this event enabled! So, you think you’re turning the tracing off for all sessions and go home, but really some sessions keep on tracing – until the filesystem is full (and you’ll get a phone call at 3am).
 

So, to be safe, you should use DBMS_MONITOR for your SQL Tracing needs, it doesn’t have the abovementioned problems. For other events you should use DBMS_SYSTEM.SET_EV/READ_EV (or ORADEBUG EVENT/SESSION_EVENT &  EVENTS/EVENTDUMP) together with ALTER SYSTEM for making sure you actually do enable/disable the events for all existing sessions too. Or better yet, stay away from undocumented events ;-)
 

If you wonder what/where is the “system event array”, it’s just a memory location in shared pool. It doesn’t seem to be explicitly visible in V$SGASTAT in Oracle 10g, but in 11.2.0.3 you get this:

No system-wide events set:

SQL> @sgastat eventPOOL         NAME                            BYTES------------ -------------------------- ----------shared pool  DBWR event stats array            216shared pool  KSQ event description            8460shared pool  Wait event pointers               192shared pool  dbgdInitEventGrp: eventGr         136shared pool  event classes                    1552shared pool  event descriptor table          32360shared pool  event list array to post           36shared pool  event list to post commit         108shared pool  event statistics per sess     2840096shared pool  event statistics ptr arra         992shared pool  event-class map                  4608shared pool  ksws service events             57260shared pool  latch wait-event table           2212shared pool  standby event stats              1216shared pool  sys event stats                539136shared pool  sys event stats for Other       32256shared pool  trace events array              7200017 rows selected.

Let’s set a system-wide event:

SQL> ALTER SYSTEM SET events = '942 TRACE NAME ERRORSTACK LEVEL 3'; System altered.

And check V$SGASTAT again:

SQL> @sgastat eventPOOL         NAME                            BYTES------------ -------------------------- ----------shared pool  DBWR event stats array            216shared pool  KSQ event description            8460shared pool  Wait event pointers               192shared pool  dbgdInitEventG                   4740shared pool  dbgdInitEventGrp: eventGr         340shared pool  dbgdInitEventGrp: subHeap          80shared pool  event classes                    1552shared pool  event descriptor table          32360shared pool  event list array to post           36shared pool  event list to post commit         108shared pool  event statistics per sess     2840096shared pool  event statistics ptr arra         992shared pool  event-class map                  4608shared pool  ksws service events             57260shared pool  latch wait-event table           2212shared pool  standby event stats              1216shared pool  sys event stats                539136shared pool  sys event stats for Other       32256shared pool  trace events array              7200019 rows selected.

 

 

So, the “system event array” lives in shared pool, as a few memory allocations with name like “dbgdInitEventG%”. Note that this naming was different in 10g, as the dbgd module showed up in Oracle 11g, when Oracle guys re-engineered the whole diagnostics event infrastructure, making it much more powerful, for example allowing you to enable dumps and traces only for a specific SQL_ID.