为什么oracle 10.2.0.5只会获取child#=1的shared pool latch

2024-04-02 19:04:59 739人浏览薄情痞子

摘要

这篇文章主要介绍了为什么oracle 10.2.0.5只会获取child#=1的shared pool latch，具有一定借鉴价值，感兴趣的朋友可以参考下，希望大家阅读完这篇文章之后大有收获，下面让小编带

这篇文章主要介绍了为什么oracle 10.2.0.5只会获取child#=1的shared pool latch，具有一定借鉴价值，感兴趣的朋友可以参考下，希望大家阅读完这篇文章之后大有收获，下面让小编带着大家一起了解一下。

1,可以用oradebug dump heapdump 3转储共享池的结构信息
这个级别一般3即可，6的代价有些大了
2,语法如下：
sql> oradebug setmypid
Statement processed.
SQL> oradebug dump heapdump 3
Statement processed.
SQL> oradebug tracefile_name
/home/ora10g/admin/ora10g/udump/ora10g_ora_6533.trc

3，转储共享池的TRC文件结构如下：
第一部分：LATCH信息
KGH Latch Directory InfORMation
ldir state: 2 Last allocated slot: 77
Slot [ 1] Latch: 0xa4222c98 Index: 2 Flags: 3 State: 2 next: (nil)

第二部分：HEAP信息，可见共计5个堆，对应子池的个数，由参数_kghdsidx_count控制
HEAP DUMP heap name="sga heap" desc=0x60000058
extent sz=0x47c0 alt=216 het=32767 rec=9 flg=-126 opc=0
parent=(nil) owner=(nil) nex=(nil) xsz=0x160
ds for latch 1: 0x60034fe0 0x60036838 0x60038090 --可见保护其子堆需要3个latch
ds for latch 2: 0x6003e808 0x60040060 0x600418b8
ds for latch 3: 0x60048030 0x60049888 0x6004b0e0
ds for latch 4: 0x60051858 0x600530b0 0x60054908
ds for latch 5: 0x6005b080 0x6005c8d8 0x6005e130 0x6005f988 --保护其子堆需要4个latch
reserved granule count 0 (granule size 16777216)

第三部分：上述每个堆的具体信息，而且TRC下述信息是以每个堆的子堆为基础展开的，其它子堆结构同理
HEAP DUMP heap name="sga heap(1,0)" desc=0x60034fe0
extent sz=0xfe0 alt=216 het=32767 rec=9 flg=-126 opc=0
parent=(nil) owner=(nil) nex=(nil) xsz=0x1000000
latch set 1 of 5
durations enabled for this heap
reserved granules for root 0 (granule size 16777216)

可见子堆由区构成，而区又包括多个CHUNK

第四部分是一个空闲列表的BUCKET列表
FREE LISTS:
Bucket 0 size=32
Bucket 1 size=40
Bucket 2 size=48
Bucket 3 size=56
Bucket 4 size=64
Bucket 5 size=72
Bucket 6 size=80
Bucket 7 size=88
Bucket 8 size=96
Bucket 9 size=104
中间略
Bucket 250 size=12352
Bucket 251 size=12360
Bucket 252 size=16408
Bucket 253 size=32792
Bucket 254 size=65560

也就是说管理空闲空间是由BUCKET进行管理，把可以分配或回收的CHUNK地址信息存储在对应的BUCKET中，具体要存储在哪个BUCKET中，要看CHUNK的大小，和对应的BUCKET进行匹配

第五部分：是一个预备的空间列表BUCKET列表（同第四部分理）
RESERVED FREE LISTS:
Reserved bucket 0 size=32
Reserved bucket 1 size=4400
Reserved bucket 2 size=8216
Reserved bucket 3 size=8696
Reserved bucket 4 size=8704
Reserved bucket 5 size=8712
Reserved bucket 6 size=8720
Reserved bucket 7 size=9368
Reserved bucket 8 size=9376
Reserved bucket 9 size=12352
Reserved bucket 10 size=12360
Reserved bucket 11 size=16408
Reserved bucket 12 size=32792
Reserved bucket 13 size=65560

第六部分：未PIN住的可以重建或重用的chunk列表（lru优先，关于LRU还要研究），如下包括很多CHUNK

UNPINNED RECREATABLE CHUNKS (lru first):
Chunk 0a3bd5420 sz= 56 recreate "fixed allocatio" latch=0x9e5c8db0 --CHUNK地址，大小，状态及类型，CHUNK对应的LATCH地址，经在TRC文件查找，可以和TRC文件第一部分的LATCH关联起来
Chunk 0a3bc7fb8 sz= 56 recreate "fixed allocatio" latch=0x9e5c7d10 --fixed allocatio对应x$ksmsp的ksmchcom,可以理解为CHUNK的名称
中间略
Chunk 0a3ba1a78 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3ba1848 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
SEPARATOR
Chunk 0a3bb2340 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3bb2110 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
中间略
Chunk 0a3b631e0 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62fb0 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62d80 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62b50 sz= 560 recreate "KQR PO " latch=0x9e5c7d10

第七部分：永久或持久的CHUNK列表，同上理，包括很多个CHUNK，不过这里仅一个CHUNK,且其类型为PERM，而且没有LATCH保护
PERMANENT CHUNKS:
Chunk 09e0cd000 sz= 15937536 perm "perm " alo=8424224
Permanent space = 15937536

4，共享池CHUNK的信息可以查询X$KSMSP
SQL> select addr,ksmchidx,ksmchcom,ksmchptr,KSMCHCLS,ksmchsiz,ksmchtyp,ksmchdur from x$ksmsp where ksmchcom='fixed allocatio' and ksmchsiz=56 and KSMCHCLS='recr' and ksmchptr='00000000A3BD5420';

ADDR KSMCHIDX KSMCHCOM KSMCHPTR KSMCHCLS KSMCHSIZ KSMCHTYP KSMCHDUR
---------------- ---------- ---------------- ---------------- -------- ---------- ---------- ----------
00002B0CBA8B5548 1 fixed allocatio 00000000A3BD5420 recr 56 72 2

5,如果HANG SHARED POOL LATCH,oradebug dump heapdump会HANG住
6，暂未在TRC文件找到SHARED POOL LATCH

7，上述TRC文件每个部分后面会列出对应部分可用空间总大小

测试

SQL> select * from v$version where rownum=1;

BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.2.0.5.0 - 64bi

--转储共享池shared pool
SQL> oradebug setmypid
Statement processed.
SQL> oradebug dump heapdump 3
Statement processed.
SQL> oradebug tracefile_name
/home/ora10g/admin/ora10g/udump/ora10g_ora_6533.trc

--TRC文件

---第一部分是一些latch的信息
KGH Latch Directory Information
ldir state: 2 Last allocated slot: 77
Slot [ 1] Latch: 0xa4222c98 Index: 2 Flags: 3 State: 2 next: (nil)
Slot [ 2] Latch: 0xa4222d78 Index: 3 Flags: 3 State: 2 next: (nil)
Slot [ 3] Latch: 0x6000a6c0 Index: 4 Flags: 3 State: 2 next: (nil)
中间略
Slot [ 75] Latch: 0x600270b0 Index: 1 Flags: 3 State: 2 next: 0x600e85c0
Slot [ 76] Latch: 0x6002abf0 Index: 2 Flags: 3 State: 2 next: (nil)
Slot [ 77] Latch: 0x60031378 Index: 3 Flags: 3 State: 2 next: 0x600e81b8

---第二部是heap的信息，可见共计5个heap堆（注： _kghdsidx_count=5，堆即分配内存的一种内存结构）
HEAP DUMP heap name="sga heap" desc=0x60000058
extent sz=0x47c0 alt=216 het=32767 rec=9 flg=-126 opc=0
parent=(nil) owner=(nil) nex=(nil) xsz=0x160
ds for latch 1: 0x60034fe0 0x60036838 0x60038090 --可见保护其子堆需要3个latch
ds for latch 2: 0x6003e808 0x60040060 0x600418b8
ds for latch 3: 0x60048030 0x60049888 0x6004b0e0
ds for latch 4: 0x60051858 0x600530b0 0x60054908
ds for latch 5: 0x6005b080 0x6005c8d8 0x6005e130 0x6005f988 --保护其子堆需要4个latch
reserved granule count 0 (granule size 16777216)

第三部分是上述每个子堆的具体信息，仅讲述一个子堆即可，其它同理

为什么oracle 10.2.0.5只会获取child#=1的shared pool latch

可知：
1，前4个堆，每个堆有3个子堆
最后一个堆，有4个子堆

2，TRC文件的下面内容是以每个堆的子堆为基础进行，我分析也以此为准

下面详解第三部分，即第一个堆的第一个子堆，即sga heap(1,0),其中1表示第一个堆，0表示第一个子堆
HEAP DUMP heap name="sga heap(1,0)" desc=0x60034fe0
extent sz=0xfe0 alt=216 het=32767 rec=9 flg=-126 opc=0
parent=(nil) owner=(nil) nex=(nil) xsz=0x1000000
latch set 1 of 5
durations enabled for this heap
reserved granules for root 0 (granule size 16777216)

可见子堆下面是一个区extent
EXTENT 0 addr=0x9e000000

可见区extent下面是很多个chunk
Chunk 09e000058 sz= 48 R-freeable "reserved stoppe" --每个chunk包括地址，大小，状态及类型
Chunk 09e000088 sz= 839496 R-free " "
Chunk 09e0ccfd0 sz= 48 R-freeable "reserved stoppe"
Chunk 09e0cd000 sz= 15937536 perm "perm " alo=8424224
Total heap size = 16777128 --这个推大小，就是上面所有chunk的大小之和

可见有一个空闲可用的列表，记录很多个bucket,每个bucket的编号及大小，共计254个bucket
FREE LISTS:
Bucket 0 size=32
Bucket 1 size=40
Bucket 2 size=48
Bucket 3 size=56
Bucket 4 size=64
Bucket 5 size=72
Bucket 6 size=80
Bucket 7 size=88
Bucket 8 size=96
Bucket 9 size=104
中间略
Bucket 250 size=12352
Bucket 251 size=12360
Bucket 252 size=16408
Bucket 253 size=32792
Bucket 254 size=65560

Total free space = 0

接着是一个预备的空闲可用的列表，格式同上，也是记录很多个bucket
RESERVED FREE LISTS:
Reserved bucket 0 size=32
Reserved bucket 1 size=4400
Reserved bucket 2 size=8216
Reserved bucket 3 size=8696
Reserved bucket 4 size=8704
Reserved bucket 5 size=8712
Reserved bucket 6 size=8720
Reserved bucket 7 size=9368
Reserved bucket 8 size=9376
Reserved bucket 9 size=12352
Reserved bucket 10 size=12360
Reserved bucket 11 size=16408
Reserved bucket 12 size=32792
Reserved bucket 13 size=65560

上述区extent中的chunk中的未使用过的chunk,注意后面的 " "
Chunk 09e000088 sz= 839496 R-free " "
而且可见chunk的信息是记录在每个bucket中

标明上述预备的空闲可用空间的大小为839496，刚好就是上述哪个chunk
Total reserved free space = 839496

未PIN住的可以重建或重用的chunk列表（lru优先，关于LRU还要研究），如下包括很多CHUNK
UNPINNED RECREATABLE CHUNKS (lru first):
Chunk 0a3bd5420 sz= 56 recreate "fixed allocatio" latch=0x9e5c8db0 --CHUNK地址，大小，状态及类型，CHUNK对应的LATCH地址，经在TRC文件查找，可以和TRC文件第一部分的LATCH关联起来
Chunk 0a3bc7fb8 sz= 56 recreate "fixed allocatio" latch=0x9e5c7d10 --fixed allocatio对应x$ksmsp的ksmchcom,可以理解为CHUNK的名称
中间略
Chunk 0a3ba1a78 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3ba1848 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
SEPARATOR
Chunk 0a3bb2340 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3bb2110 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
中间略
Chunk 0a3b631e0 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62fb0 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62d80 sz= 560 recreate "KQR PO " latch=0x9e5c7d10
Chunk 0a3b62b50 sz= 560 recreate "KQR PO " latch=0x9e5c7d10

标明上述未PIN住的空间大小
Unpinned space = 221984 rcr=78 trn=322

永久或持久的CHUNK列表，同上理，包括很多个CHUNK，不过这里仅一个CHUNK,且其类型为PERM，而且没有LATCH保护
PERMANENT CHUNKS:
Chunk 09e0cd000 sz= 15937536 perm "perm " alo=8424224
Permanent space = 15937536

标明上述永久的CHUNK空间的大小

我们继续分析
--x$ksmsp记录共享池中chunk的相关信息，可见共计20917个CHUNK
SQL> select count(*) from x$ksmsp;

COUNT(*)
----------
20917

查询上述未PIN住的可以重建或重用的chunk列表第一个CHUNK
SQL> select addr,ksmchidx,ksmchcom,ksmchptr,KSMCHCLS,ksmchsiz,ksmchtyp,ksmchdur from x$ksmsp where ksmchcom='fixed allocatio' and ksmchsiz=56 and KSMCHCLS='recr' and ksmchptr='00000000A3BD5420';

ADDR KSMCHIDX KSMCHCOM KSMCHPTR KSMCHCLS KSMCHSIZ KSMCHTYP KSMCHDUR
---------------- ---------- ---------------- ---------------- -------- ---------- ---------- ----------
00002B0CBA8B5548 1 fixed allocatio 00000000A3BD5420 recr 56 72 2

由下可见TRC文件第一部分LATCH对应V$LATCH_chidlren,且注意：ADDR为小写，不要用大写，否则查询不到信息
SQL> select addr,latch#,level#,name from v$latch_children where lower(addr) like '%a4222c98%';

ADDR LATCH# LEVEL# NAME
---------------- ---------- ---------- --------------------------------------------------
00000000A4222C98 29 0 ksfv messages

但是仍然找不到shared pool latch

加大DUMP级别看看，可否找到shared pool latch
SQL> oradebug setmypid
Statement processed.
SQL> oradebug dump heapdump 10
ORA-00085: current call does not exist
SQL> oradebug dump heapdump 6
Statement processed.
SQL> oradebug tracefile_name
/home/ora10g/admin/ora10g/udump/ora10g_ora_8143.trc

还是找不到shared pool latch,转换思路，先HANG shared pool latch,再查看DUMP文件，看可否有，如还没有，就是我分析思路不对
SQL> oradebug setmypid
Statement processed.
SQL> oradebug poke 0x00000000600E7AF0 4 1
BEFORE: [0600E7AF0, 0600E7AF4) = 00000000
AFTER: [0600E7AF0, 0600E7AF4) = 00000001
SQL> oradebug setmypid
Statement processed.

不过好现如果HANG SHARED POOL LATCH，发现oradebug dump heapdump 6 也hang住了
SQL> oradebug dump heapdump 6

只能以PRELIM方式先恢复SHARED POOL LATCH
[ora10g@seconary ~]$ sqlplus -prelim '/as sysdba'

SQL*Plus: Release 10.2.0.5.0 - Production on Thu Nov 19 07:37:53 2015

Copyright (c) 1982, 2010, Oracle. All Rights Reserved.

SQL> oradebug poke 0x00000000600E7AF0 4 0
ORA-00074: no process has been specified
SQL> oradebug setmypid
Statement processed.
SQL> oradebug poke 0x00000000600E7AF0 4 0
BEFORE: [0600E7AF0, 0600E7AF4) = 000000FF
AFTER: [0600E7AF0, 0600E7AF4) = 00000000

发现heapdump 3也不行会HANG
SQL> oradebug setmypid
Statement processed.
SQL> oradebug dump heapdump 3
Statement processed.

这样，HANG住CHILD#=2的shared pool latch,看什么情况,最后发现也会HANG住，可能因为不是一个子池的原因，深入原因还要研究
SQL> oradebug setmypid
Statement processed.
SQL> oradebug poke 0x00000000600E7B90 4 1
BEFORE: [0600E7B90, 0600E7B94) = 00000000
AFTER: [0600E7B90, 0600E7B94) = 00000001