转 -- Oracle Interested Transaction List (ITL) Waits

原址如下：

http://www.rampant-books.com/art_nanda_interested_tarnsaction_list_itl.htm

OracleInterested Transaction List (ITL) Waits

by Arup Nanda

Author of Oracle Privacy Security Auditing

What is ITL? Ever wondered how Oracle locks rows on behalf oftransactions? In some RDBMS vendor implementations, a lock manager maintainsinformation on which row is locked by which transaction. This works great intheory, but soon the lock manager becomes a single point of contention, as eachtransaction must wait to get a lock from the manager and then wait again torelease the lock.

This severely limits the scalability ofthe applications. In fact, application developers of some RDBMS productsdespise holding locks for a long time, and often resort to a full table lockwhen all that's needed is to get a few rows locked. This creates further waits,and consequently, scalability suffers.

So how is that different in Oracle? Forstarters, there is no lock manager. When a row is locked by a transaction, thatinformation is placed in the block header where the row is located. Whenanother transaction wishes to acquire the lock on the same row, it has totravel to the block containing the row anyway, and upon reaching the block, itcan easily tell that the row is locked from the block header. There is no needto queue up for some single resource like a lock manager. This makesapplications immensely scalable.

So, what portion of the block headercontains information on locking? It is a simple data structure called"Interested Transaction List" (ITL), a linked list data structurethat maintains information on transaction address and rowid. ITL containsseveral slots or place holders for transactions. When a row in the block islocked for the first time, the transaction places a lock in one of the slotswith the rowid of the row that is locked. In other words, the transaction makesit known that it is interested in the row (hence the name "InterestedTransaction List").

When the same transaction or another onelocks another row, the information is stored in another slot, and so on. Aftera transaction ends via commit or a rollback, the locks are released and so arethe slots that were used to mark the blocks, and these newly freed slots arereused for the other transactions. So there is in fact a queue, but it's at ablock level, not at the entire database level or even at a segment level.

The next logical question that comes upis, how many slots are typically available? During the table creation, theINITRANS parameter defines how many slots are initially created in the ITL.When the transactions exhaust all the available slots and a new transactioncomes in to lock a row, the ITL grows to create another slot. The ITL can growup to the number defined by the MAXTRANS parameter of the table, provided thereis space in the block. Nevertheless, if there is no more room in the block,even if the MAXTRANS is high enough, the ITL cannot grow.

So, what happens when a transaction doesnot find a free slot to place its lock information? This can occur becauseeither (i) the block is so packed that the ITL cannot grow to create a freeslot, or (ii) the MAXTRANS has already been reached. In this case, thetransaction that needs to lock a row has to wait until a slot becomesavailable. This wait is termed as ITL waits and can be seen from the viewv$session_wait, in which the session is waiting on an event named"enqueue." Since the INITRANS is one, there is only one slotfor the ITL. The rest of the block is empty.

Then another transaction, Txn2, updates the row Row2 and wants to lock the row.However, there are no more slots in the ITL available to service thetransaction. The MAXTRANS entry is 11, meaning the ITL can grow up to 11 slotsand the block has empty space. Therefore, ITL can grow by another slot and Slotnumber two is created and allocated to Txn2 (refer to figure 4).

Now the empty space in the block isseverely limited, and it will not be able to fit another ITL slot. If at thistime another transaction comes in to update the row three, it must have a freeslot in the ITL. The MAXTRANS is 11 and currently only two slots have beencreated, so another one is possible; but since there is no room in the block togrow, the slot can't be created. Therefore, the Txn3 has to wait until eitherof the other transactions rolls back or commits and the slot held by it becomesfree. At this time the session will experience an ITL waits event as seen fromthe view V$SESSION_WAIT.

To better illustrate the concept, let'sillustrate such waits using a case. Create the following table and thenpopulate it with several rows. Note MAXTRANS value.

CREATE TABLE TAB1
( COL1 NUMBER,
COL2 VARCHAR2(200))
INITRANS 1 MAXTRANS 1
/

DECLARE
I NUMBER;
BEGIN
FOR I IN 1..10000 LOOP
INSERT INTO TAB1 VALUES
(I,'SSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSS');
END LOOP;
COMMIT;
END;
/

Now update a rowof the table from one session, but do not commit it.
UPDATE TAB1 SET COL2 = 'UPDATED' WHERE COL1 = 1;

From another session, update row number two and do not update it.
UPDATE TAB1 SET COL2 = 'UPDATED' WHERE COL1 = 2;

This session will wait. Why? It's updating a row for COL1 = 2, not the same rowupdated in the other session for COL1 = 1. So why is the session waiting? It'sbecause the first transaction occupied the only available ITL slot. The secondtransaction needed another slot to place its lock information, but since theMAXTRANS I defined is one, the ITL could not grow to create another slot. Thus,the latter transaction has to wait until the former session releases the lockand makes the slot available. Now increase the MAXTRANS of the table byissuing

ALTER TABLE TAB1 MAXTRANS 11;

and redo the above test. The second session will not wait this time because theITL had enough free slots for both transactions. How to Reduce ITL WaitsTheprimary cause of ITL waits is that free slots in the ITL are not available.This can be due to low setting of the MAXTRANS, which places a hard limit onthe number of transactions that can have locks on a block or, the block is sopacked that there is no room for the ITL to grow OR or both.

Therefore, setting a high value of INITRANS will make sure that there areenough free slots in the ITL, and there will be minimal or no dynamic extensionof the ITL. However, doing so also means that there is less space in the blockfor actual data, increasing wasted space.
The other option is to making sure the data is less packed so that ITL can growenough to accommodate the surges in ITL. This can be done by increasingPCTFREE, increasing FREELISTS and FREELIST GROUPS parameters for a table.

This will make ablock hold less data and more room for the ITL to grow. As a direct result ofthe reduction in packing, the table will experience fewer buffer busy waitevents, and performance will be increased.
How to Diagnose the ITL WaitHow do you know that a segment is experiencing ITLwaits? The best answer will be found in the Segment Level Statistics providedin Oracle9i Release 2. To check for ITL waits, set up the STATISTICS_LEVEL toTYPICAL in init.ora or via ALTER SYSTEM, then examine the segment statisticsfor the waits.

SELECT
   OWNER,
   OBJECT_NAME
FROM
   V$SEGMENT_STATISTICS
WHERE
   STATISTIC_NAME = 'ITL waits'
AND
   VALUE > 0;

This unearths the objects that were subjected to ITL waits since the start upof the instance. Note that this view resets when the instance bounces. (For amore detailed explanation of this view and how to set it up, please refer tothe article by this author here in DBAzine.)
In versions earlier than 9i, checking for ITL waits is tricky.

When you suspectthat a database is suffering from these waits, query the view v$session_wait.If the event on which the system is waiting is "enqueue," then thesession might be experiencing ITL waits. However, enqueue is a very broad eventthat encompasses any type of locks, so it does not accurately specify the ITLwaits. When the wait event is experienced, issue the following complex query:

Selects.sid SID,
s.serial# Serial#,
l.type type,
' ' object_name,
lmode held,
request request
from v$lock l, v$session s, v$process p
where s.sid = l.sid and
s.username <> ' ' and
s.paddr = p.addr and
l.type <> 'TM' and
(l.type <> 'TX' or l.type = 'TX' and l.lmode <> 6)
union
select s.sid SID,
s.serial# Serial#,
l.type type,
object_name object_name,
lmode held,
request request
from v$lock l, v$session s, v$process p, sys.dba_objects o
where s.sid = l.sid and
o.object_id = l.id1 and
l.type = 'TM' and
s.username <> ' ' and
s.paddr = p.addr
union
select s.sid SID,
s.serial# Serial#,
l.type type,
'(Rollback='||rtrim(r.name)||')' object_name,
lmode held,
request request
from v$lock l, v$session s, v$process p, v$rollname r
where s.sid = l.sid and
l.type = 'TX' and
l.lmode = 6 and
trunc(l.id1/65536) = r.usn and
s.username <> ' ' and
s.paddr = p.addr
order by 5, 6
/

The output of the query will look something like this:

SID SERIAL# TYOBJECT_NAM HELD REQUEST
----- ------- -- ---------- ---------- --------
36 8428 TX 0 4
36 8428 TM TAB1 3 0
52 29592 TM TAB1 3 0
52 29592 TX (Rollback=RBS1_6) 6 0

Note how the sessions 36 and 52 both have a TM (DML) lock on the table TAB1 oftype 3 (Row Exclusive), but session 52 also holds a TX (Transaction) lock onthe rollback segment of mode 6 (Exclusive) and Session 36 is waiting for a mode4 (Share) lock. If this combination of locking occurs, you can be sure thatsession 36 is waiting for ITL on the table TAB1. Beware of a similar butdifferent diagnosis when two sessions try to insert the same key value (a reallocking – primary key violation). In that case, you would also see anadditional TX lock on a rollback segment from the session that is waiting; forITL waits, this TX lock on the rollback segment would not be seen.

Needless to say, this is a rather convoluted and inaccurate way to diagnose theITL waits in pre-Oracle9i Release 2 versions.

What INITRANS Value is Optimal? Conversely, how do you know if the INITRANSsetting is too high and the space is just being wasted? Ascertaining this ispossible by using a few random block dumps from the segment in question. First,find out the header file# and header block# of the segment by issuing thefollowing query:

SELECTHEADER_FILE, HEADER_BLOCK FROM DBA_SEGMENTSWHERE OWNER = '...'AND SEGMENT_NAME= '...';

Use the output of the query to do a block dump of the header block.
ALTER SYSTEM DUMP DATAFILE <file#> BLOCK MIN <block#> BLOCK MAX<block#>;

This will produce a trace file in the USER_DUMP_DESTINATION directory. Open thetrace file and find out the section on extent control via the following:
Extent Control Header

-----------------------------------------------------------------

Extent Header::spare1: 0 spare2: 0 #extents: 1 #blocks: 10

last map 0x00000000 #maps: 0 offset: 2080

Highwater:: 0x02011f87 ext#: 0 blk#: 0 ext size: 10

#blocks in seg. hdr's freelists: 0

#blocks below: 0

mapblk 0x00000000 offset: 0

Unlocked

Map Header:: next 0x00000000 #extents: 1 obj#: 53689 flag: 0x40000000

Extent Map

-----------------------------------------------------------------

0x02011f87 length: 10

Find out the real number of blocks for the segment from dba_segments via thefollowing:
SELECT BLOCKS FROM DBA_SEGMENTS

WHERE OWNER = '...' AND SEGMENT_NAME = '...';

Say this returns 12, and the #blocks shows 10; this means the first two blocksare header blocks; the data starts at the third block. Take a dump of the thirdblock, which is obtained by adding two to the header block# obtained above.
ALTER SYSTEM DUMP DATAFILE <file#> BLOCK MIN <block#> BLOCK MAX<block#>;

This will produce another trace file in the USER_DUMP_DEST directory. If youissued it during the same session as above, then the trace will be written inthe trace file opened earlier. Open the file and locate the following section.
buffer tsn: 8 rdba: 0x02011f88 (8/73608)

scn:0x0000.389b8d81 seq: 0x01 flg: 0x04 tail: 0x8d810601

frmt: 0x02 chkval: 0x2ef5 type: 0x06=trans data

Block header dump: 0x02011f88

Object id on Block? Y

seg/obj: 0xd1ad csc: 0x00.389b8d81 itc: 4 flg: - typ: 1 - DATA

fsl: 0 fnx: 0x0 ver: 0x01

Itl Xid Uba Flag Lck Scn/Fsc

0x01 0x0003.003.000024cc 0x00804067.050a.13 C-U- 0 scn 0x0000.389b304e

0x02 0x0007.010.00002763 0x00801f49.0453.01 C--- 0 scn 0x0000.389b2628

0x03 0x0002.00a.000025d5 0x00804d42.04b2.25 C--- 0 scn 0x0000.389b2811

0x04 0x0006.006.00002515 0x00800962.03c8.18 CU 0 scn 0x0000.389b3044

This shows some very important information on the block, especially in the ITLsection shown above. This table has an INITRANS entry of four, so there arefour lines, one each per the ITL. The Flag column above the flag -U- indicatesthat the particular ITL was used. In this case, only two of the ITLs were used,and the other two were never used. However, this is the case for this blockonly. By selecting block dumps from other randomly selected blocks, you couldhave an idea how many ITLs are actually used. Then you may decide to reduce theINITRANS.

Automatic Block Management in Oracle9iIn Oracle9i, the process of space mangementinside a block is somewhat changed due to the introduction of the AutomaticBlock Management (ABM) feature, also known as Automatic Segment SpaceManagement (ASSM). The option is specified at the tablespace level in thestorage parameter as SEGMENT SPACE MANAGEMENT AUTO. For instance, thetablespace TS1 can be created as

CREATE TABESPACETS1
DATAFILE '...'
EXTENT MANAGEMENT LOCAL

SEGMENT SPACE MANAGEMENT AUTO;The last line of this code does the magic. In theABM mode, Oracle maintains a bitmap for each segment with the information onthe block. A bitmap is a data structure with a bit representing each block.When a block becomes available for INSERT, the information is made availablesimply by setting the corresponding bit in the bitmap rather than usingfreelists.

So, what does thishave to do with ITL waits? The very cause of ITL waits is not freespacemanagement, but the unavailability of a slot in ITL waits. So you still have tolook for ITL waits and correct them using INITRANS and MAXTRANS. In fact, theproblem may become exacerbated because the block becomes quite packed followingan efficient space management system, and that may lead to lack of space forITL growth. You can prevent this by keeping a large INITRANS for the segment.

Proper setting ofINITRANS and MAXTRANS and packing of the blocks is vital to avoid ITL waits inOracle. It's interesting to note that locking doesn't cause waits, but rather,the mechanism for locking as well as and poor planning. However, the good newsis that this situation can be easily fixed by reorganizing the table and addingmore slots to the Interested Transaction List.

转 -- Oracle Interested Transaction List (ITL) Waits相关推荐

什么是Interested Transaction List(ITL)--Oracle？
Oracle ITL. 1.什么是Interested Transaction List(ITL)? 2.什么是Interested Transaction List(ITL)等待? 3.怎样减少In ...
Interested Transaction List(ITL)
在以前的RDBMS系统中,很多进行事物操作的时候,对行都是进行表锁.在oracle中就不再有锁管理器.当行被事物锁住的时候,相关锁信息放在行所在的块头中.当其他的事物想获得相同块的锁时,无论怎么样都会 ...
what is ITL( Interested Transaction list)
what is ITL( Interested Transaction list) Segment Transaction Slot Internals Interested Transaction ...
Oracle Block浅析2：ITL(Interested Transaction List)
一.ITL(Interested Transaction List): ITL(Interested Transaction List)是Oracle数据块内部的一个组成部分,位于数据块头(block ...
Oracle ITL(Interested Transaction List)理解
ITL(Interested Transaction List) ITL是位于数据块头部的事物槽列表,它是由一系列的ITS(Interested Transaction Slot,事物槽)组成,其初始 ...
ITL(Interested Transaction List)理解
一.ITL描述: ITL(Interested Transaction List)是Oracle数据块内部的一个组成部分,位于数据块头(block header),itl由xid,uba,flag,l ...
Orace ITL(Interested Transaction List) 说明
一. ITL 说明 ITL: Interested Transaction List,也叫事务槽,它位于BLOCK Header.先dump 一个block,看一下这个ITL的信息. SELECTa. ...
zt_Orace ITL(Interested Transaction List) 说明
http://blog.csdn.net/tianlesoftware/article/details/6573988 来自 " ITPUB博客 " ,链接:http://blog ...
关于Oracle AUTONOMOUS TRANSACTION(自治事务)的介绍
AUTONOMOUS TRANSACTION(自治事务)的介绍在基于低版本的ORACLE做一些项目的过程中,有时会遇到一些头疼的问题,比如想在执行当前一个由多个DML组成的transaction(事 ...

转 -- Oracle Interested Transaction List (ITL) Waits

转 -- Oracle Interested Transaction List (ITL) Waits相关推荐

最新文章

热门文章