cache buffers chains以及热块解决方案

cache buffers chains以及热块解决方案

今天是2013-10-10，今天下午我调休了，中午饭过后从14点一直睡到16点，这种感觉真爽。
 之前学习过关于buffer cache的cache buffers lru chain的知识，其中说到了lru，dirty list，ws等等，详见：http://blog.csdn.net/xiaohai20102010/article/details/11490925

今天在看一下buffer bucket 与 cache buffers chain
为了快速找到查询中所需要的block是否在cache buffer中，oracle划分了bucket，根据每个数据块dba进行hash计算，然后存放到不同的bucket中区，bucket中存有cache buffers chain以及相关block的buffer header，举个例子，buffer相当于一个图书馆，bucket相当于每个抽屉，在这个抽屉中记录了是每个buffer中块的信息也就是buffer header，这个名片的信息（buffer header）通过cache buffers chain进行串联起来，每个抽屉都要有专门人进行管理，不然大家都翻卡片找自己的书那么不乱了嘛，这个管理这就是cache buffer chain latch当我们找书的时候，首先经过hash计算找到buffer位于哪个bucket中，然后通过获得cache buffers chain lath然后从cache buffers chain上 拿到每个buffer的名片信息，然后再去获得这个buffer，如果没有找到，那么就会从数据文件中去读这个所需信息。（通俗易懂的说发，呵呵)
 ORACLE 中可以通过查看隐含参数可以知道有多个bucket，就像查看x$KCBWDS可以知道有多少个ws一样。
eg：

SQL> set verify off
SQL> @getsp.sql
Enter value for par: db_block_hash

KSPPINM                                            KSPPSTVL             KSPPDESC
-------------------------------------------------- -------------------- ------------------------------------------------------------
_db_block_hash_buckets                             65536                Number of database block hash buckets
_db_block_hash_latches                             2048                 Number of database block hash latches

SQL>
可以看到有65536个bucket，但是latch的数量为：2048，也就是说每个cache buffers chain latch 管理32个bucket，65536/2048=32,在google找到了一个图，如下，可以更加明白：

首先我通过转储buffers然后找到一个chain如下：
EG:
ALTER SYSTEM SET EVENTS 'IMMEDAITE TRACE NAME BUFFERS LEVEL 10';

CHAIN: 35 LOC: 0x7bcf8bb0 HEAD: [0x733dc118,0x733dc118]
    BH (0x733dc060) file#: 1 rdba: 0x004075b5 (1/30133) class: 1 ba: 0x730a0000
      set: 3 pool: 3 bsz: 8192 bsi: 0 sflg: 2 pwc: 475,28
      dbwrid: 0 obj: 225 objn: 225 tsn: 0 afn: 1 hint: f
      hash: [0x7bcf8bb0,0x7bcf8bb0] lru: [0x733dc288,0x733dc018]
      lru-flags: debug_dump
      ckptq: [NULL] fileq: [NULL] objq: [0x733dc790,0x733dbf08] objaq: [0x733dc7a0,0x733dbf18]
      st: XCURRENT md: NULL fpin: 'kdswh05: kdsgrp' tch: 2
      flags:
      LRBA: [0x0.0.0] LSCN: [0x0.0] HSCN: [0xffff.ffffffff] HSUB: [65535]
      buffer tsn: 0 rdba: 0x004075b5 (1/30133)
      scn: 0x0000.0002d9f1 seq: 0x01 flg: 0x06 tail: 0xd9f10601
      frmt: 0x02 chkval: 0x1ca5 type: 0x06=trans data
Hex dump of block: st=0, typ_found=1
可以看到这是第35个chain，其中包含的是buffer header 信息，bh就是代表buffer header的，在其中还可以找到刚刚提及到的hash计算，hash: [0x7bcf8bb0,0x7bcf8bb0]代表下一个buffer header 的地址以及上一个buffer header地址，lru: [0x733dc288,0x733dc018]代表在lru中下一个buffer块的位和上一个lru中buffer块的位置。另外还有一个tch：2，这个代表这个快被两次访问，如果该值越到，那么这个块就越热手 ，大家都在找她，也就会出现热点块的概念，同时伴随着chache buffers chain的latch 竞争。
对于chain中的bh的信息，在oracle中还存有一个fixed table 那就是x$bh，对于该视图介绍，请看博客地址：
http://blog.csdn.net/xiaohai20102010/article/details/10285449
这里介绍几个常用的字段信息：
TS#         NUMBER 8.X Tablespace number 
DBARFIL     NUMBER 8.X Relative file number of block 
DBAFIL      NUMBER 7.3 File number of block 
DBABLK      NUMBER Block number of block 
NXT_HASH    NUMBER Next buffer on this hash chain 
PRV_HASH    NUMBER Previous buffer on this hash chain 
NXT_LRU     NUMBER Next buffer on the LRU 
PRV_LRU     NUMBER Previous buffer on the LRU 
HLADDR      RAW(4) Hash Chain Latch Address 
TCH     NUMBER 8.1 Touch Count
next_hash以及PRV_HASH 对应于上文提到的hash: [0x7bcf8bb0,0x7bcf8bb0]，NXT_LRU以及PRV_LRU对应上文提到的lru: [0x733dc288,0x733dc018]，tch对应于上文提到的tch: 2。等等等等，还有很多信息反正bufferheader的信息都在x$bh中可以找到。但是注意oracle一直在变动这些表，具体字段还需依据系统版本来定。

select ts#,dbarfil,dbafil,dbablk,nxt_hash,prv_hash,nxt_lru,prv_lru,hladdr,tch from x$bh where nxt_hash='0x7bcf8bb0'；

一、
知道如上信息了，首先查看一下数据库中，最热手的数据块信息；
eg：
SQL> select * from (select ts#,file#,dbarfil,dbablk,state,tch,nxt_hash,prv_hash,nxt_repl,prv_repl from x$bh order by tch desc) where rownum<20;

TS#      FILE#    DBARFIL     DBABLK      STATE        TCH NXT_HASH         PRV_HASH         NXT_REPL         PRV_REPL
---------- ---------- ---------- ---------- ---------- ---------- ---------------- ---------------- ---------------- ----------------
         0          1          1       2017          1        255 000000007BDC00B0 000000007BDC00B0 0000000073FF9DD8 0000000073FF65C8
         0          1          1       2016          1        254 000000007BE481C0 000000007BE481C0 0000000071FDE370 0000000071FDD740
         1          2          2      62997          1        235 000000007BE20D10 0000000070FD82F0 0000000073FF7BB8 000000006FFEADC0
         1          2          2      39076          1        201 000000007BCFD820 000000007BCFD820 000000006FFF1690 000000006FFF12E8
         1          2          2      36844          1        196 000000007BDC69E0 000000007BDC69E0 000000006FFEF6E0 000000006FFEF0C8
         1          2          2      36839          1        150 000000007BDA68F0 00000000777F7CB8 00000000777FB500 00000000777FA660
         1          2          2      39131          1        126 000000007BCFA8B0 000000007BCFA8B0 000000006FFF07F0 000000006FFF0580
         1          2          2      39127          1        126 000000007BDB66B0 000000007BDB66B0 000000006FFF06B8 000000006FFF0448
         1          2          2      60461          1        126 000000007BDB0550 000000007BDB0550 000000006FFF01D8 000000006FFEFF68
         1          2          2      36826          1        126 000000007BD9A400 000000007BD9A400 000000006FFEFE30 000000006FFEFBC0
         0          1          1      66493          1        126 00000000727E3618 000000007BD99600 000000006FFF0CD0 000000006FFF0A60

TS#      FILE#    DBARFIL     DBABLK      STATE        TCH NXT_HASH         PRV_HASH         NXT_REPL         PRV_REPL
---------- ---------- ---------- ---------- ---------- ---------- ---------------- ---------------- ---------------- ----------------
         0          1          1      77698          1        126 000000007BD95EB0 000000006F3E4D40 000000006FFF0928 000000006FFF06B8
         0          1          1        465          1        126 000000007BD81070 000000007BD81070 000000006FFF0B98 000000006FFF0928
         1          2          2      39083          1        126 0000000072FE7578 000000007BD71D30 0000000073FFAA08 000000006FFF0CD0
         1          2          2      60459          1        126 000000007BD5C130 000000007BD5C130 000000006FFF0448 000000006FFF01D8
         1          2          2      36824          1        126 000000007BD45FE0 000000007BD45FE0 000000006FFEFBC0 000000006FFEF950
         1          2          2      39073          1        126 0000000072FF72F8 000000007BD31B50 000000006FFF12E8 000000006FFF1558
         0          1          1        497          1        126 000000007BD31A30 000000007BD31A30 000000006FFF0E08 000000006FFF0B98
         1          2          2      60462          1        126 000000007BD28440 000000007BD28440 000000006FFF00A0 000000006FFEFE30

19 rows selected.

SQL>
可以看到在dbablk为：2017，它的nxt_hash与prv_hash都是一个，也就是说这个块自己在一个chain上，这个chain中在没有其他数据块。

知道了这个信息，我们可以知道这些热块存在于那个表空间，那个数据文件，那个数据块是热的等等信息。

二、
根据x$bh找到访问频繁的对象。可以与dba_extents关联

eg：
SQL>  select owner,segment_name,partition_name,segment_type,tablespace_name from dba_extents a,
  2   (select * from
  3      (select ts#,file#,dbarfil,dbablk,state,tch,nxt_hash,prv_hash,nxt_repl,prv_repl from x$bh order by tch desc)
  4    where rownum<20) b
  5    where a.file_id=b.file# and a.block_id <= b.dbablk and a.block_id+a.blocks >= b.dbablk and a.relative_fno=b.dbarfil;
 
 
OWNER                          SEGMENT_NAME                                                                     PARTITION_NAME                 SEGMENT_TYPE       TABLESPACE_NAME
------------------------------ -------------------------------------------------------------------------------- ------------------------------ ------------------ ------------------------------
SYS                            JOB$                                                                                                            TABLE              SYSTEM
SYS                            JOB$                                                                                                            TABLE              SYSTEM
SYS                            SCHEDULER$_CLASS                                                                                                TABLE              SYSTEM
SYS                            SCHEDULER$_JOB                                                                                                  TABLE              SYSTEM
SYS                            SCHEDULER$_JOB                                                                                                  TABLE              SYSTEM
SYS                            SCHEDULER$_JOB                                                                                                  TABLE              SYSTEM
SYS                            SCHEDULER$_LIGHTWEIGHT_JOB                                                                                      TABLE              SYSTEM
SYS                            C_USER#                                                                                                         CLUSTER            SYSTEM
SYS                            C_USER#                                                                                                         CLUSTER            SYSTEM
SYS                            I_USER#                                                                                                         INDEX              SYSTEM
SYS                            I_USER1                                                                                                         INDEX              SYSTEM
SYS                            I_SYSAUTH1                                                                                                      INDEX              SYSTEM
SYS                            I_SYSAUTH1                                                                                                      INDEX              SYSTEM
SYS                            I_SYSAUTH1                                                                                                      INDEX              SYSTEM
SYS                            I_SYSAUTH1                                                                                                      INDEX              SYSTEM
SYS                            I_SYSAUTH1                                                                                                      INDEX              SYSTEM
SYS                            I_PROFILE                                                                                                       INDEX              SYSTEM
SYS                            I_JOB_JOB                                                                                                       INDEX              SYSTEM
SYS                            I_JOB_NEXT                                                                                                      INDEX              SYSTEM
SYS                            SCHEDULER$_CLASS_PK                                                                                             INDEX              SYSTEM
 
OWNER                          SEGMENT_NAME                                                                     PARTITION_NAME                 SEGMENT_TYPE       TABLESPACE_NAME
------------------------------ -------------------------------------------------------------------------------- ------------------------------ ------------------ ------------------------------
SYS                            SCHEDULER$_CLASS_PK                                                                                             INDEX              SYSTEM
SYS                            SCHEDULER$_LOBJ_UK                                                                                              INDEX              SYSTEM
 
22 rows selected
 
SQL>
三、
根据x$bh 也可以找到对应于获得该cache buffers chain 的latch信息，注意不能与v$latch进行关联，因为该视图是个汇总，需要与v$latch_children视图进行关联：
eg：
SQL> select name,gets,misses,sleeps,immediate_gets,immediate_misses,spin_gets,wait_time
  2  from v$latch_children a,
  3  (select * from
  4   (select ts#,file#,dbarfil,dbablk,state,tch,nxt_hash,hladdr,prv_hash,nxt_repl,prv_repl from x$bh order by tch desc)
  5   where rownum<20) b
  6    where a.ADDR=b.hladdr and a.name='cache buffers chains' order by gets desc;
 
NAME                                                                   GETS     MISSES     SLEEPS IMMEDIATE_GETS IMMEDIATE_MISSES  SPIN_GETS  WAIT_TIME
---------------------------------------------------------------- ---------- ---------- ---------- -------------- ---------------- ---------- ----------
cache buffers chains                                                  94752          0          0             71                0          0          0
cache buffers chains                                                  37077          0          0             60                0          0          0
cache buffers chains                                                  29415          0          0             70                0          0          0
cache buffers chains                                                  27973          0          0             67                0          0          0
cache buffers chains                                                  21447          0          0             71                0          0          0
cache buffers chains                                                  15893          0          0             64                0          0          0
cache buffers chains                                                  10095          0          0             66                0          0          0
cache buffers chains                                                   8585          0          0            146                0          0          0
cache buffers chains                                                   3782          0          0             76                0          0          0
cache buffers chains                                                   3744          0          0             67                0          0          0
cache buffers chains                                                   3447          0          0             70                0          0          0
cache buffers chains                                                   3386          0          0             65                0          0          0
cache buffers chains                                                   2785          0          0             76                0          0          0
cache buffers chains                                                   2782          0          0             71                0          0          0
cache buffers chains                                                   2683          0          0             67                0          0          0
cache buffers chains                                                   2488          0          0             69                0          0          0
cache buffers chains                                                   2443          0          0             66                0          0          0
cache buffers chains                                                   2300          0          0             62                0          0          0
cache buffers chains                                                   1895          0          0             71                0          0          0
 
19 rows selected
 
SQL>
四、
如果存在cache buffers chains 可以通过statspack report查看得到信息，可以参考v$session_wait 视图找到free latch 等待事件，并且可以通过如下语句找到相应的latch 地址，然后我们在去组合找到对象，也是一种方法
eg：
查找属于cache buffers chains的latch信息，注意下面是演示，没有真正出现cache buffers chains latch竞争现象。
SQL> select addr,name,gets,misses,sleeps,immediate_gets,immediate_misses,spin_gets,wait_time
  2     from v$latch_children where name='cache buffers chains' and rownum<20 order by sleeps desc
  3  ;
 
ADDR             NAME                                                                   GETS     MISSES     SLEEPS IMMEDIATE_GETS IMMEDIATE_MISSES  SPIN_GETS  WAIT_TIME
---------------- ---------------------------------------------------------------- ---------- ---------- ---------- -------------- ---------------- ---------- ----------
000000007BE5B278 cache buffers chains                                                   1243          0          0             72                0          0          0
000000007BE58468 cache buffers chains                                                    915          0          0             71                0          0          0
000000007BE5B0E8 cache buffers chains                                                   9720          0          0             69                0          0          0
000000007BE5B020 cache buffers chains                                                    957          0          0             65                0          0          0
000000007BE5AF58 cache buffers chains                                                    924          0          0             69                0          0          0
000000007BE5AE90 cache buffers chains                                                    514          0          0             67                0          0          0
000000007BE5ADC8 cache buffers chains                                                   1220          0          0             71                0          0          0
000000007BE5AD00 cache buffers chains                                                   1120          0          0             66                0          0          0
000000007BE59C38 cache buffers chains                                                   1041          0          0            139                0          0          0
000000007BE59B70 cache buffers chains                                                    473          0          0             66                0          0          0
000000007BE59AA8 cache buffers chains                                                    899          0          0             70                0          0          0
000000007BE599E0 cache buffers chains                                                    851          0          0             66                0          0          0
000000007BE59918 cache buffers chains                                                   3122          0          0             67                0          0          0
000000007BE59850 cache buffers chains                                                    487          0          0             65                0          0          0
000000007BE59788 cache buffers chains                                                    738          0          0             77                0          0          0
000000007BE596C0 cache buffers chains                                                   1008          0          0             69                0          0          0
000000007BE585F8 cache buffers chains                                                    756          0          0             69                0          0          0
000000007BE58530 cache buffers chains                                                   1983          0          0             72                0          0          0
000000007BE5B1B0 cache buffers chains                                                    539          0          0             63                0          0          0
 
19 rows selected
 
SQL>
既然找到了latch的addr，那么也就可以找到相应的对象信息：
eg：
SQL>   select b.ts#,b.file#,b.dbarfil,b.dbablk,b.tch,a.gets,a.misses,a.sleeps,a.immediate_gets,a.immediate_misses from
  2    (select * from
  3    (select addr,ts#,file#,dbarfil,dbablk,hladdr,tch,nxt_hash,prv_hash,nxt_repl,prv_repl from x$bh
  4     order by tch desc) where rownum<20 )b,
  5     (select addr,name,gets,misses,sleeps,immediate_gets,immediate_misses,spin_gets,wait_time
  6   from v$latch_children where name='cache buffers chains' order by misses desc) a
  7    where a.addr=b.hladdr order by b.tch desc;
 
       TS#      FILE#    DBARFIL     DBABLK        TCH       GETS     MISSES     SLEEPS IMMEDIATE_GETS IMMEDIATE_MISSES
---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- -------------- ----------------
         0          1          1      12498        175       9117          0          0            151                0
         0          1          1      12584        172       4000          0          0             71                0
         0          1          1      12496        172       4048          0          0             82                0
         0          1          1      12481        172       3666          0          0             68                0
         0          1          1      12489        172       2105          0          0             77                0
         0          1          1      12497        170       3029          0          0             83                0
         0          1          1        964        168      99290          0          0             76                0
         0          1          1       2025        166       3789          0          0             72                0
         0          1          1        963        166      38918          0          0             60                0
         0          1          1        965        166      29381          0          0             71                0
         1          2          2      36771        151      16726          0          0             68                0
         0          1          1      77728        151       2506          0          0             68                0
         1          2          2      38275        151       2897          0          0             70                0
         0          1          1        417        111       5778          0          0             70                0
         1          2          2      36847         89       2964          0          0             72                0
         0          1          1      46461         77       7481          0          0             88                0
         0          1          1      12097         76       2741          0          0             70                0
         0          1          1      12145         76       3046          0          0             80                0
         0          1          1      65330         76      12240          0          0             68                0
 
19 rows selected
 
SQL>
五、
如下是看到一个blog，在此学习一下。
 
热点块和热点对象我们都找到了，但是我们该怎么来解决这个问题呢？一般来说，热点块会导致cache buffers chains竞争等待，但并不是说cache buffer chains一定是因为热点块而起，在特别情况下有可能是因为latch数量的问题导致的，也就是一个latch管理的buffers数量太多而导致竞争激烈。但是latch数量我们一般是不会轻易去设置的，这是oracle的隐藏参数。
 
实际上最有效的办法，是从优化sql入手，不良的sql往往带来大量的不必要的访问，这是造成热点块的根源。比如本该通过全表扫描的查询却走了索引的range scan，这样将带来大量的对块的重复访问。从而形成热点问题。再或者比如不当地走了nested loops的表连接，也可能对非驱动表造成大量的重复访问。那么在这个时候，我们的目标就是找出这些sql来并尝试优化。在statspack报告中，根据报告中sql列表，我们如果是通过dba_extents确定的热点对象而不是通过dba_objects确定的，则可以通过查找出的热点segment转换为对应的表，对于非分区的索引，index_name就是segment_name,通过dba_indexes很容易的找到对应的table_name,对于分区表和分区索引也能通过和dba_tab_partition和dba_ind_partitions找到segment和table的对应关系。通过这些table到statspack报告中去找相关的sql。
 
select sql_text
from stats$sqltext a,
(select distinct a.owner,a.segment_name,a.segment_type from
dba_extents a,
(select dbarfil,dbablk
from (select dbarfil,dbablk
from x$bh order by tch desc) where rownum < 11) b
where a.RELATIVE_FNO = b.dbarfil
and a.BLOCK_ID <= b.dbablk and a.block_id + a.blocks > b.dbablk) b
where a.sql_text like '%'||b.segment_name||'%' and b.segment_type = 'TABLE'
order by a.hash_value,a.address,a.piece;