[20260505]关于内核参数kernel.shmmax.txt
--//在探究ipcs -m 输出shmid值遇到的问题，单独写一篇blog。
1.环境:
# cat /etc/redhat-release
CentOS Linux release 7.3.1611 (Core)
SYS@book> @ ver2
==============================
PORT_STRING : x86_64/Linux 2.4.xx
VERSION : 21.0.0.0.0
BANNER : Oracle Database 21c Enterprise Edition Release 21.0.0.0.0 - Production
BANNER_FULL : Oracle Database 21c Enterprise Edition Release 21.0.0.0.0 - Production
Version 21.3.0.0.0
BANNER_LEGACY : Oracle Database 21c Enterprise Edition Release 21.0.0.0.0 - Production
CON_ID : 0
PL/SQL procedure successfully completed.
2.测试：
SYS@book> startup nomount force
ORACLE instance started.
Total System Global Area 1107294056 bytes
Fixed Size 9684840 bytes
Variable Size 654311424 bytes
Database Buffers 436207616 bytes
Redo Buffers 7090176 bytes
# ipcs -m --human
------ Shared Memory Segments --------
key shmid owner perms size nattch status
0x00000000 0 oracle 600 10M 34
0x00000000 32769 oracle 600 1G 34
0x00000000 65538 oracle 600 8M 34
0xafa94c20 98307 oracle 600 2M 34
--//显示1G，实际上是1090519040=1040M。
# ipcs -m
------ Shared Memory Segments --------
key shmid owner perms bytes nattch status
0x00000000 0 oracle 600 10485760 34
0x00000000 32769 oracle 600 1090519040 34
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
0x00000000 65538 oracle 600 8388608 34
0xafa94c20 98307 oracle 600 2097152 34
--//4个共享内存段，想人为拆分建立7个共享内存段。这样需要将1090519040/1024/1024 = 1040M分成4份。
--//1040/4 = 260M, 这样建立每个300M看看,换算成字节位 300*1024*1024 = 314572800。
3.继续：
# sysctl -w kernel.shmmax=$((300*1024*1024))
kernel.shmmax = 314572800
SYS@book> startup nomount force
ORACLE instance started.
Total System Global Area 1107294056 bytes
Fixed Size 9684840 bytes
Variable Size 654311424 bytes
Database Buffers 436207616 bytes
Redo Buffers 7090176 bytes
# ipcs -m --human
------ Shared Memory Segments --------
key shmid owner perms size nattch status
0x00000000 131072 oracle 600 10M 34
0x00000000 163841 oracle 600 288M 34
0x00000000 196610 oracle 600 288M 34
0x00000000 229379 oracle 600 288M 34
0x00000000 262148 oracle 600 176M 34
0x00000000 294917 oracle 600 8M 34
0xafa94c20 327686 oracle 600 2M 34
# ipcs -m
------ Shared Memory Segments --------
key shmid owner perms bytes nattch status
0x00000000 131072 oracle 600 10485760 34
0x00000000 163841 oracle 600 301989888 34
0x00000000 196610 oracle 600 301989888 34
0x00000000 229379 oracle 600 301989888 34
0x00000000 262148 oracle 600 184549376 34
0x00000000 294917 oracle 600 8388608 34
0xafa94c20 327686 oracle 600 2097152 34
--//建立7个共享内存段.
--//301989888/1024/1024 = 288M,实际上拆分的共享内存段选择的并不是300M，而是288+288+288+176 = 1040.
--//实际上上午的测试选择260M,实际上拆分的共享内存段选择的256M,这样拆分成5个段,加起来建立8个共享内存段.
--//没有满足测试需求,感觉应该与Granule Size大小相关。
SYS@book> select * from v$sgainfo where name = 'Granule Size';
NAME BYTES RESIZE CON_ID
---------------------------------------- ---------- ------ ----------
Granule Size 16777216 No 0
--//猜测应该按照16倍数来选择共享段大小。这样如果按16倍数是256，272，288，304，这样选择304正好拆分.
# sysctl -w kernel.shmmax=$((304*1024*1024))
kernel.shmmax = 318767104
--//SYS@book> startup nomount force
# ipcs -m --human
------ Shared Memory Segments --------
key shmid owner perms size nattch status
0x00000000 360448 oracle 600 10M 34
0x00000000 393217 oracle 600 304M 34
0x00000000 425986 oracle 600 304M 34
0x00000000 458755 oracle 600 304M 34
0x00000000 491524 oracle 600 128M 34
0x00000000 524293 oracle 600 8M 34
0xafa94c20 557062 oracle 600 2M 34
--//验证正确。
4.顺便测试最小kernel.shmmax等于多少。
# sysctl -w kernel.shmmax=$((16*1024*1024))
kernel.shmmax = 16777216
--//SYS@book> startup nomount force
# ipcs -m --human
------ Shared Memory Segments --------
key shmid owner perms size nattch status
0x00000000 589824 oracle 600 10M 34
0x00000000 622593 oracle 600 16M 34
0x00000000 655362 oracle 600 16M 34
0x00000000 688131 oracle 600 16M 34
0x00000000 720900 oracle 600 16M 34
0x00000000 753669 oracle 600 16M 34
0x00000000 786438 oracle 600 16M 34
0x00000000 819207 oracle 600 16M 34
0x00000000 851976 oracle 600 16M 34
0x00000000 884745 oracle 600 16M 34
0x00000000 917514 oracle 600 16M 34
0x00000000 950283 oracle 600 16M 34
0x00000000 983052 oracle 600 16M 34
0x00000000 1015821 oracle 600 16M 34
0x00000000 1048590 oracle 600 16M 34
0x00000000 1081359 oracle 600 16M 34
0x00000000 1114128 oracle 600 16M 34
0x00000000 1146897 oracle 600 16M 34
0x00000000 1179666 oracle 600 16M 34
0x00000000 1212435 oracle 600 16M 34
0x00000000 1245204 oracle 600 16M 34
0x00000000 1277973 oracle 600 16M 34
0x00000000 1310742 oracle 600 16M 34
0x00000000 1343511 oracle 600 16M 34
0x00000000 1376280 oracle 600 16M 34
0x00000000 1409049 oracle 600 16M 34
0x00000000 1441818 oracle 600 16M 34
0x00000000 1474587 oracle 600 16M 34
0x00000000 1507356 oracle 600 16M 34
0x00000000 1540125 oracle 600 16M 34
0x00000000 1572894 oracle 600 16M 34
0x00000000 1605663 oracle 600 16M 34
0x00000000 1638432 oracle 600 16M 34
0x00000000 1671201 oracle 600 16M 34
0x00000000 1703970 oracle 600 16M 34
0x00000000 1736739 oracle 600 16M 34
0x00000000 1769508 oracle 600 16M 34
0x00000000 1802277 oracle 600 16M 34
0x00000000 1835046 oracle 600 16M 34
0x00000000 1867815 oracle 600 16M 34
0x00000000 1900584 oracle 600 16M 34
0x00000000 1933353 oracle 600 16M 34
0x00000000 1966122 oracle 600 16M 34
0x00000000 1998891 oracle 600 16M 34
0x00000000 2031660 oracle 600 16M 34
0x00000000 2064429 oracle 600 16M 34
0x00000000 2097198 oracle 600 16M 34
0x00000000 2129967 oracle 600 16M 34
0x00000000 2162736 oracle 600 16M 34
0x00000000 2195505 oracle 600 16M 34
0x00000000 2228274 oracle 600 16M 34
0x00000000 2261043 oracle 600 16M 34
0x00000000 2293812 oracle 600 16M 34
0x00000000 2326581 oracle 600 16M 34
0x00000000 2359350 oracle 600 16M 34
0x00000000 2392119 oracle 600 16M 34
0x00000000 2424888 oracle 600 16M 34
0x00000000 2457657 oracle 600 16M 34
0x00000000 2490426 oracle 600 16M 34
0x00000000 2523195 oracle 600 16M 34
0x00000000 2555964 oracle 600 16M 34
0x00000000 2588733 oracle 600 16M 34
0x00000000 2621502 oracle 600 16M 34
0x00000000 2654271 oracle 600 16M 34
0x00000000 2687040 oracle 600 16M 34
0x00000000 2719809 oracle 600 16M 34
0x00000000 2752578 oracle 600 8M 34
0xafa94c20 2785347 oracle 600 2M 34
--//2785347 = 0x2a8043,0x43+1 = 68，建立68个共享内存段。
--//顺便验证上午确定shmid的计算公式是否正确,确定下次启动ipcs -m 的shmid值:
2^16 * (B+S/2)
--//B:前次相乘因子 开始0,S 前次启动共享内存段数量, 开始0.
--//589824: 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3
--//B = 9 ,S=68
2^16 * (B+S/2) = 2^16 * (9+68/2) = 2818048
# sysctl -w kernel.shmmax=4398046511104
kernel.shmmax = 4398046511104
--//SYS@book> startup nomount force
# ipcs -m
------ Shared Memory Segments --------
key shmid owner perms bytes nattch status
0x00000000 2818048 oracle 600 10485760 34
0x00000000 2850817 oracle 600 1090519040 34
0x00000000 2883586 oracle 600 8388608 34
0xafa94c20 2916355 oracle 600 2097152 34
--//2818048,与猜测一致.
--//2818048: 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 43
5.继续测试:
# sysctl -w kernel.shmmax=$((8*1024*1024))
kernel.shmmax = 8388608
SYS@book> startup nomount force
ORA-12547: TNS:lost contact
--//无法在会话执行startup nomount force,连接已经断开.
--//只能重新登录,但是报ORA-12547: TNS:lost contact,kernel.shmmax只要小于16M,无法登录.
$ sqlplus -s -l / as sysdba
ERROR:
ORA-12547: TNS:lost contact
SP2-0751: Unable to connect to Oracle. Exiting SQL*Plus
$ oerrz ORA-12547
12547, 00000, "TNS:lost contact"
// *Cause: Partner has unexpectedly gone away, usually during process
// startup.
// *Action: Investigate partner application for abnormal termination. On an
// Interchange, this can happen if the machine is overloaded.
--//这个提示能把dba带偏.
# sysctl -w kernel.shmmax=$((16*1024*1024))
kernel.shmmax = 16777216
$ sqlplus / as sysdba
SQL*Plus: Release 21.0.0.0.0 - Production on Tue May 5 16:10:43 2026
Version 21.3.0.0.0
Copyright (c) 1982, 2021, Oracle. All rights reserved.
Connected to an idle instance.
SYS@book>
--//也就是最小设置kernel.shmmax = 16777216,这样与Granule Size也没有关系.
# sysctl -w kernel.shmmax=$((15*1024*1024))
kernel.shmmax = 15728640
--//即使停止监听，也是报一样的错误。
$ lsnrctl stop
LSNRCTL for Linux: Version 21.0.0.0.0 - Production on 05-MAY-2026 16:24:54
Copyright (c) 1991, 2021, Oracle. All rights reserved.
Connecting to (DESCRIPTION=(ADDRESS=(PROTOCOL=TCP)(HOST=192.168.56.101)(PORT=1521)))
The command completed successfully
$ sqlplus -s -l / as sysdba <<<quit
ERROR:
ORA-12547: TNS:lost contact
SP2-0751: Unable to connect to Oracle. Exiting SQL*Plus
--//如果这样定位问题需要一定难度,不会想到kernel.shmmax设置太小.
$ strace -f -y -o test9.txt sqlplus -s -l / as sysdba <<<quit
ERROR:
ORA-12547: TNS:lost contact
SP2-0751: Unable to connect to Oracle. Exiting SQL*Plus
$ grep kernel test9.txt
5037 open("/proc/sys/kernel/shmmax", O_RDONLY) = 8
5037 fstat(8</proc/sys/kernel/shmmax>, {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
5037 read(8</proc/sys/kernel/shmmax>, "15728640\n", 1024) = 9
5037 close(8</proc/sys/kernel/shmmax>) = 0
5038 open("/proc/sys/kernel/shmmax", O_RDONLY) = 4
5038 fstat(4</proc/sys/kernel/shmmax>, {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
5038 read(4</proc/sys/kernel/shmmax>, "15728640\n", 1024) = 9
5038 close(4</proc/sys/kernel/shmmax>) = 0
--//涉及到内核参数仅仅取出kernel.shmmax值。
