转自老白的<oracle rac 日记>一书,
dump systemstate产生的跟踪文件包含了系统中所有进程的进程状态等信息。每个进程对应跟踪文件中的一段内容,反映该进程的状态信息,包括进程信息,会话信息,enqueues信息(主要是lock的信息),缓冲区的信息和该进程在SGA区中持有的(held)对象的状态等信息。dump systemstate产生的跟踪文件是从dump那一刻开始到dump任务完成之间一段事件内的系统内所有进程的信息。
那么通常在什么情况下使用systemstate比较合适呢?
Oracle推荐的使用systemstate事件的几种情况是:
数据库hang住了
数据库很慢
进程正在hang
资源争用
dump systemstate的语法为:
ALTER SESSION SET EVENTS ‘immediate trace name systemstate level 10’;
也可以使用ORADEBUG实现这个功能:
sqlplus -prelim / as sysdba
oradebug setmypid
oradebug unlimit;
oradebug dump systemstate 10
如果希望在数据库发生某种错误时调用systemstate事件,可以在参数文件(spfile或者pfile)中设置event参数,
例如,当系统发生死锁(出现ORA-00060错误)时dump systemstate:
event = “60 trace name systemstate level 10”
LEVEL参数:
10Dump all processes (IGN state)
5Level 4 + Dump all processes involved in wait chains (NLEAF state)
4Level 3 + Dump leaf nodes (blockers) in wait chains (LEAF,LEAF_NW,IGN_DMP state)
3Level 2 + Dump only processes thought to be in a hang (IN_HANG state)
1-2Only HANGANALYZE output,no process dump at all
如果Level过大的话会产生大量的跟踪文件并影响系统的I/O性能,建议不要采用3级以上的跟踪。Hanganalyze报告会分作许多片断,会话片断信息总是由一个header详尽描述被提取的的会话信息。
一般来说,一份systemstate dump中包含了以下内容:
dump header文件头
process dump dump时所有的process的dump信息,每个process一个专门的章节。
call dump在process dump中,包含call dump
session dump每个process中,都有1个或多个(MTS时)session dump
enqueue dump
buffer dump在session dump中可能包含buffer dump
在阅读systemstate dump时,一般首先使用ASS工具来进行分析。ASS是oracle工程师编写的一个AWK脚本,用于分析systemstate dump文件,找出dump中可能存在问题的地方。通过ASS的输出结果,我们就可以发现一些blocker的线索,这些线索就是我们重点要查看的地方。
我们可以通过搜索SO的地址信息来定位某个SO,找到后分析这个SO的信息,并且通过PARENT SO的地址找到其PARENT,建立这些SO的关系图。比如我们找到一个SESSION的SO,就可以看看这个session属于哪个process,这个session正在执行的sql是什么,等等。通过这种分析,就把可能存在问题的SO及关联的SO全部找出来,这样就为进一步分析提供了素材。
1、标准的state object header(SO)
state object header中包含了一些基本的信息,比如:
SO: c00004ti4jierj,type: 2. owner: 0000000000,flag: init/-/-/0x00
其中SO是state object的号码;
type表示state object的类别;
TYPE: state object的已知类别:
2 process(进程)
3 call
4 session(会话信息)
5 enqueue(锁信息)
6 file infomation block(文件信息块,每个FIB标识一个文件)
11 broadcast handle(广播消息句柄)
12 KSV slave class state
13 ksvslvm
16 osp req holder(会话执行os操作的holder)
18 enqueue resource detail(锁资源详细资料)
19 ges message(ges消息)
20 namespace [ksxp] key
24 buffer [db buffer]
36 dml lock
37 temp table lock(临时表锁)
39 list of blocks(用于block cleanout的块列表清单)
40 transaction(事务)
41 dummy
44 sort segment handle(排序段句柄)
50 row cache enqueue
52 user lock
53 library cache lock
54 library cache pin
55 library cache load lock
59 cursor enqueue
61 process queue
62 queue reference
75 queue monitor sob
owner是这个SO的父节点(如果为0,说明是最顶层的SO);flag表示状态,值有以下三种:
kssoinit;state object被初始化了
kssoflst;state object在freelist上
kssofcln;state object已经被pmon释放了。
State object header的数据结构如下:
struct kssob {
unsigned charkssobtyp; /* state object的类别*/
unsigned charkssobflg; /* flags */
unsigned charkssobdelstage;
struct kssob *kssobown; /*拥有者的SO指针*/
kgglkkssoblnk; /*在父对象成员链中的指针*/
}
2、processstate dump(ksupr)
processstate dump转储了进程的状态,从这些信息中我们可以了解进程的基本属性以及进程的状态。
在阅读processstate dump时,我们主要关注的进程的标识(FLAG),从中也可以知道进程的类别。从”(latch info)”中可以看到进程等待latch的情况,这也有助于了解进程故障的原因。另外,进程的OS信息对于进一步了解进程情况也是很有帮助的。
实际上,x$ksupr包含了进程的信息,通过该内存视图可以更进一步了解processstate dump的内容。
ADDR地址
INDX序号
INST_ID实例ID
KSSPAFLGstate object的状态:
KSSOINIT 0x01 // state object initialized
KSSOFLST 0x02 // state object is on free list
KSSOFCLN 0x04 // state object freed by PMON(for debugging)
KSSPAOWN该SO的OWNER,如果自己是顶层的SO,那么owner为0
KSUPRFLG该process的状态:
KSUPDEAD 0x01process is dead and should be cleaned up
KSUPDSYS 0x02detached,system process
KSUPDFAT 0x04detached,fatal(system) process
KSUPDCLN 0x08process is cleanup(pmon)
KSUPDSMN 0x10process is smon
KSUPDPSU 0x20pseudo process
KSUPDMSS 0x40muti-stated server
KSUPDDPC 0x80dispatcher process
KSUPRSER进程的序号(SERIAL NO)
KSUPRIOC
KSLLALAQ持有的latch
KSLLAWAT正在等待的latch
KSLLAWHYlatch请求的上下文(用于debug)
KSLLAWERlatch请求的位置(用于debug)
KSLLASPN本进程正在spin的latch
KSLLALOW所持有latch级别的位图(0~9级)
KSLLAPSC进程发出的POST消息的计数
KSLLAPRC进程收到的POST消息的计数
KSLLAPRV收到的最后一个POST的LOC ID,参考图中的①
KSLLAPSN最后一个发送POST的LOC ID,参考图中的②
KSLLID1RRESOURCE ID的第一部分
KSLLID2RRESOURCE ID的第二部分
KSLLRTYPRESOURCE TYPE+RESOURCE FLAG
KSLLRMTYRESOURCE MANAGE的类型:
KRMENQ0x01enqueues
KRMLATCH0x02latches
KRMLIBCALK 0x03library cache locks
KRMBUFLK0x04buffer locks
KSLLARPO最后一个发送消息给这个进程的OS进程
KSLLASPO这个进程最后一个发送信息过去的OS进程
KSUPRPIDOS进程号
KSUPRWID等待事件的ID
KSUPRUNMOS用户名
KSUPRMNM用户的机器名
KSUPRPNM用户程序名
KSUPRTID用户终端名
KSSRCOBJSTATE OBJECT RECOVERY数据中的正在被操作的对象
KSSRCFRESTATE OBJECT RECOVERY数据中的FREELIST的地址
KSSRCSRCSTATE OBJECT RECOVERY数据中的SOURCE PARENT
KSSRCDSTSTATE OBJECT RECOVERY数据中的DESTINATION PARENT
KSASTQNXMESSAGE STATE中的前向指针
KSASTQPRMESSAGE STATE中的后向指针
KSASTRPLMESSAGE STATE中的REPLY VALUE
KSUPRPGPPROCESS GROUP的名字
KSUPRTFI进程的trace文件名
KSUPRPUMPGA使用的内存
KSUPRPNAM KSUPRPNAM+KSUPRPRAM是pga分配内存的总和
KSUPRPRAM
KSUPRPFMpga可释放的内存
KSUPRPMMpga使用的最大内存
3、session state object
会话信息中包含了大量我们所需要的信息,一般来说会话状态块是我们分析会话情况的重点。
在会话状态信息中,flag是十分重要的,我们可以从flag中了解会话目前的情况,以及flag位图的详细信息。该会话正在执行的sql和pl/sql的SO地址可以让我们找到当前会话正在做的工作,有助于进一步
分析。另外,会话的等待事件和历史等待事件可以让用户了解会话在现在和过去一段时间里等待的情况,如果要分析会话故障原因的话,这些资料都是十分重要的。
flag的位图如下:
KSUSFUSR0x00000001user session (as opposed to recursive session)
KSUSFREC0x00000002recursive session(always internal)
KSUSFAUD0x00000004audit logon/logoff,used by cleanup
KSUSFDCO0x00000008disable commit/rollback from plsql
KSUSFSYS0x00000010user session created by system processes
KSUSFSGA0x00000020whether UGA is allocate in sga
KSUSFLOG0x00000040whether user session logs on to ORACLE
KSUSFMSS0x00000080user session created by multi-stated server
KSUSFDIT0x00000100disable (defer) interrupt
KSUSFCLC0x00000200counted for current license count decrement
KSUSFDET0x00000400session has been detached
KSUSFFEX0x00000800“forced exit”during shutdown normal
KSUSFCAC0x00001000(cloned) session is cached
KSUSFILS0x00002000default tx isolation level is serializable
KSUSFOIL0x00004000override serializable for READ COMMITTED
KSUSFIDL0x00008000idle session scheduler
KSUSFSKP0x00010000SKIP unusable indexes maintenance
KSUSFCDF0x00020000defer all deferrable constraint by default
KSUSFCND0x00040000deferable constraints are immediate
KSUSFIDT0x00080000session to be implicitly detached
KSUSFTLA0x00100000transaction audit logged
KSUSFJQR0x00200000recource checking in job q process enabled
KSUSFMGS0x00400000session is migratable
KSUSFGOD0x00800000migratable session need to get ownership id
KSUSFSDS0x01000000suppress/enable TDSCNcomputations
KSUSFMSP0x02000000parent of migratable session
KSUSFMVC0x04000000MV container update progress
KSUSFNAS0x08000000an NLS alter session call was done
KSUSFTRU 0x10000000 a trusted callout was performed
KSUSFHOA0x20000000an HO agent was called
KSUSFSTZ0x40000000an alter session set time_zone was done
KSUSFSRF0x80000000summary refresh
4、call state object
Call state object是针对一个call的,我们查看call state object的时候一定要注意depth值,以此判断该call是用户调用还是递归调用。
5、enqueue state object
从enqueue state object中,我们主要可以查看锁的类型、锁的模式以及flag。
6、transaction dump
Transaction dump对应的oracle内存结构是KTCXB,可以通过X$KTCXB来了解更详细的情况。
flag的描述如下(资料来源早期版本,针对10g可能略有不同):
1allocated but no transaction
2transaction active
4state object no longer valid
8transaction about to commit/abort
10space management transaction
20recursive transaction
40no undo logging
80no change/commit,must rollback
100use system undo segment (0)
200valid undo segment assigned
400undo seg assigned,lock acquired
800change may have been made
1000assigned undo seg
2000required lock in cleanup
4000is a pseudo space extent
8000save the tx table & tx ctl block
10000no read-only optimize for 2pc
20000multiple sess attached to this tx
40000commit scn future set
80000dependent scn future set
100000dist call Failed,force rollback
200000remote uncoordinated ddl tx
400000coordinated global tx
800000pdml transaction
1000000next must be commit or rollback
2000000coordinator in pdml
4000000disable block level recovery
8000000library and/or row caches dirty
10000000serializable transaction
20000000waiting for unbound transaction
40000000loosely coupled transaction branch
80000000long-running transaction
flag2的描述如下(资料来源早期版本,针对10g可能略有不同):
1tx needs refresh on commit
2delete performed in tx
4concurrency check enabled
8insert performed
10dir path insert performed
20fast rollback on net disconnect
40do not commit this tx
80this tx made remote change
100all read-only optim enabled
事务环境的结构如下:
Struct ktcev {
kenvktcevenv;
kubaUBA的高水位;
kubaktcevucl;
sb2在undo高水位块中的剩余空间;
kcbdsundo block的描述;
kdbafrundo段头的DBA地址;
kturt *指向undo seg的KTURT结构;
7、library object lock/handle
library object lock如下:
Flags的描述如下:
KGLLKBRO0x0100this lock is broken
KGLLKCBB0x0200this lock can be broken
KGLLKPNC0x0400“kgllkpnc” is a valid pin for the call
KGLLKPNS0x0800“kgllkpns” is a valid pin for the session
KGLLKCGA0x1000this lock is in CGA memory
KGLLKINH0x2000the instance lock is inherited
KGLLKLRU0x4000lock protects an object on the session cache lru
KGLLKKPC0x8000lock protects an object in the session keep cache
KGLLKRES0x0010reserved lock preventing handle from being freed
KGLLKCBK0x0020need to callback the client for delete/dump
作为library object的主体,handle的信息如图:
其中namespace的取值包括:
CRSRcursor
TABLtable/view/sequence/synonym
BODYbody(e.g.,package body)
TRGRtrigger
INDXindex
CLSTcluster
KGLTinternal KGL testing
PIPEpipe
LOB lob
DIRdirectory
QUEUqueue
OBJGreplication object group
PROPreplication propagator
JVSCjava source
JVREjava resource
ROBJreserved for server-side RepAPI
REIPreplication internal package
CPOBcontext policy object
EVNTpub_sud internal information
SUMMsummary
DIMNdimension
CTXapp context
OUTLstored outlines
TULSruleset objects
RMGRresource manager
XDBSxdb schema
PPLNpending scheduler plan
PCLSpending scheduler class
SUBSsubscription information
LOCSlocation information
RMOBremote objects info
RSMDRepAPI snapshot Metadata
JVSDjava shared data
STFGfile group
TRANStransformation
RELCreplication � log based child
STRMstream:capture process in log-based replication
REVCrule evaluation context
STAPstream:apply process in log-based replication
RELSsource inlog-based replication
RELDdestination in log-based replication
IFSDIFS schema
XDBCXDB configuration management
USAGuser agent mapping
VOMDTABLmulti-versioned object for table
JSQIscheduler-event queue info object
CDCSchange set
VOMDINDXmulti-versioned object for index
STBOsql tuing base object
HTSPhintsetobject
JSGAscheduler global attributes
JSETscheduler start time namespace
TABL_Ttemporary table
CLST_Ttemporary cluster
INDX_Ttemporary index
SCPDsratch pad
JSLVscheduler job slave
MODLmining models
状态标志位的取值:
EXSexistent
NEXno-existent
LOC
CRTbeing created
ALTbeing altered
DRPbeing dropped
PRGbeing purged
UPDbeing uodated
RIVmarked for rolling invalidation
NRCdon’t recover when an exclusive pin fails
UDPdep being updated
BOWbad owner of database link
MEMhas frame memory associated with heap 0
REA protected with read-only access at least once
NOAprotected with no access at least once
通过对library cache object/handle的分析,可以找到相关的sql以及cursor的状态。