LMSn没有运行在RT (real time) 模式Oracle 19c RAC?

Oracle希望在数据库主机CPU使用率枯竭时，尽可能让核心的几个后台进程可以最大优先级获取CPU, 当然CPU过高会导致I/O 响应时间变长和网络延迟增加，也会间接影响数据的整体性能。 

从Oracle 10g开始是有隐藏参数_high_priority_processes控制哪些进程是高优先级,19c参数中除了_high_priority_processes，还增加了_highest_priority_processes控制高优先级。在10.2版本中，Oracle缺少_high_priority_processes对RAC的核心进程LMS*设置高优先级，在11g版本中对LMS*|VKTM设置高优先级，在19c版本中_highest_priority_processes对VKTM是最高优先级，且提供了更多对LMS*|LM*|LCK0|GCR*|CKPT|DBRM|RMS0|LGWR|CR*|RMV*配置高优先级 。

记得在10.2.0.3前有个bug会导致使用CPU进程过高。最近有客户19c RAC遇到CPU使用率超过90%时GC问题较为突出，在查看LMS时，没有在RT模式的状态引起了注意，在19c中LMS还是有一些变化，下面进行了简单的记录。

在Linux平台上，进程的内核调用模式分为三类：

• TS – SCHED_OTHER (SCHED_NORMAL) ，这是分时调度策略，缺省的通用级别；

• FF – SCHED_FIFO，这是实时调度策略，先进先出；

• RR – SCHED_RR，实时调度策略，时间片轮转;

先看一个正常环境 Oracle 19c RAC 2-nodes on RHEL 7.8

db alert log

Starting background process CLMN
CLMN started with pid=3, OS id=28714
Starting background process PSP0
PSP0 started with pid=4, OS id=28731
Starting background process IPC0
2021-03-23 10:07:32.440000 +08:00
IPC0 started with pid=5, OS id=29420
Starting background process VKTM
Starting background process GEN0
VKTM started with pid=6, OS id=29445 at elevated (RT) priority
VKTM running at (1)millisec precision with DBRM quantum (100)ms
Starting background process MMAN

Starting background process LMD1
LMD0 started with pid=23, OS id=29631
* Load Monitor used for high load check
* New Low - High Load Threshold Range = [130560 - 174080]
LMS1 started with pid=26, OS id=29640_29663 at elevated (RT) priority
LMS0 started with pid=24, OS id=29635_29662 at elevated (RT) priority
LMS2 started with pid=28, OS id=29646_29666 at elevated (RT) priority
Starting background process LMD2
LMD1 started with pid=36, OS id=29659
LMS3 started with pid=30, OS id=29649_29667 at elevated (RT) priority
LMS4 started with pid=32, OS id=29651_29672 at elevated (RT) priority
LMS5 started with pid=34, OS id=29653_29677 at elevated (RT) priority
Starting background process LMD3
LMD2 started with pid=37, OS id=29681
LMD3 started with pid=38, OS id=29686
Starting background process RMS0
RMS0 started with pid=39, OS id=29689


oracle@anbob_com:/home/oracle>  ps -efc|grep vktm
grid      34874      1 RR   41 Jun03 ?        00:06:20 asm_vktm_+ASM1
oracle    42358      1 RR   41 Jun03 ?        00:05:24 ora_vktm_anbob1
grid      58462      1 RR   41 Jun03 ?        00:06:18 mdb_vktm_-MGMTDB

Note:
使用ps-c选项查看进程优先级时， vktm是RR mode。

oracle@anbob_com:/home/oracle> ps -efc|grep lms
oracle    35148  90946 TS   19 16:02 pts/3    00:00:00 grep --color=auto lms
oracle    66573      1 TS   19 May21 ?        04:32:32 ora_lms0_anbob1
oracle    66576      1 TS   19 May21 ?        04:29:41 ora_lms1_anbob1
oracle    66578      1 TS   19 May21 ?        04:26:33 ora_lms2_anbob1
oracle    66581      1 TS   19 May21 ?        04:26:51 ora_lms3_anbob1
oracle    66586      1 TS   19 May21 ?        04:25:38 ora_lms4_anbob1
oracle    66589      1 TS   19 May21 ?        04:28:44 ora_lms5_anbob1
oracle    66596      1 TS   19 May21 ?        04:25:44 ora_lms6_anbob1
oracle    66599      1 TS   19 May21 ?        04:50:02 ora_lms7_anbob1
oracle    66603      1 TS   19 May21 ?        04:22:42 ora_lms8_anbob1
oracle    66609      1 TS   19 May21 ?        04:21:31 ora_lms9_anbob1
oracle    66615      1 TS   19 May21 ?        04:25:41 ora_lmsa_anbob1
oracle    66620      1 TS   19 May21 ?        04:29:43 ora_lmsb_anbob1
grid     129022      1 TS   19 May14 ?        00:36:49 asm_lms0_+ASM1

Note:
使用ps-c选项查看进程优先级时，lms还是TS Mode。在12c版本及之前PS也是显示RR mode，如下：

# sqlplus -V
SQL*Plus: Release 12.2.0.1.0 Production

# ps -eLfc |head -n 1;ps -eLfc|grep lms
UID        PID  PPID   LWP NLWP CLS PRI STIME TTY          TIME CMD
grid     14661     1 14661    1 RR   41  2019 ?        1-08:14:40 asm_lms0_+ASM1
oracle   62106     1 62106    1 RR   41  2019 ?        17-22:45:22 ora_lms0_weejar1
oracle   62109     1 62109    1 RR   41  2019 ?        18-10:30:26 ora_lms1_weejar1
oracle   62111     1 62111    1 RR   41  2019 ?        18-00:13:16 ora_lms2_weejar1
oracle   62113     1 62113    1 RR   41  2019 ?        17-22:02:20 ora_lms3_weejar1
oracle   62115     1 62115    1 RR   41  2019 ?        17-22:07:53 ora_lms4_weejar1

检查oradism文件

oracle@anbob_com:/home/oracle> ls -l $ORACLE_HOME/bin/oradism
-rwsr-x--- 1 root oinstall 147848 Apr 17 2019 /oracle/app/oracle/product/19c/db_1/bin/oradism

正常。
Note:
For 10gR2 and 11gR1 installations, verify that the oradism executable matches the following ownership and permissions “-rwsr-sr-x 1 root dba oradism” and make sure the lms is running in Real Time mode.

检查Oracle_HOME文件系统挂载点

oracle@anbob_com:/home/oracle> cat /proc/mounts|grep oracle
/dev/mapper/fusioncube-oracle /oracle ext4 rw,relatime,stripe=16,data=ordered 0 0

正常。

AWR中LMS

RAC Statistics
BeginEnd
Number of Instances:22
Number of LMS’s:1212
Number of realtime LMS’s:1212(0 priority changes)

检查后台进程

SQL> select 'LMS', INST_ID,PRIORITY,COUNT(*) TOTAL FROM GV$BGPROCESS where name like 'LMS%' GROUP BY INST_ID,PRIORITY ;

'LMS'     INST_ID PRIORITY              TOTAL
------ ---------- ---------------- ----------
LMS             1 RT                       12
LMS             2 RT                       12

种种显示当前LMS进程是RT模式，但PS显示进程还是TS，难道是显示问题？还是Oracle有新特性改变?

答案是的确发生了新变化，从18c开始LMS进程改为线程模式。

oracle@anbob_com:/home/oracle> ps -eLfc |head -n 1;ps -eLfc|grep lms
UID         PID   PPID    LWP NLWP CLS PRI STIME TTY          TIME CMD
oracle    66573      1  66573    4 TS   19 May21 ?        00:00:08 ora_lms0_anbob1
oracle    66573      1  66580    4 RR   41 May21 ?        03:15:29 ora_lms0_anbob1
oracle    66573      1  67219    4 TS   19 May21 ?        00:23:08 ora_lms0_anbob1
oracle    66573      1  67240    4 TS   19 May21 ?        00:53:41 ora_lms0_anbob1
oracle    66576      1  66576    4 TS   19 May21 ?        00:00:08 ora_lms1_anbob1
oracle    66576      1  66582    4 RR   41 May21 ?        03:12:36 ora_lms1_anbob1
oracle    66576      1  67270    4 TS   19 May21 ?        00:23:09 ora_lms1_anbob1
oracle    66576      1  67301    4 TS   19 May21 ?        00:53:43 ora_lms1_anbob1
oracle    66578      1  66578    4 TS   19 May21 ?        00:00:08 ora_lms2_anbob1
oracle    66578      1  66591    4 RR   41 May21 ?        03:10:10 ora_lms2_anbob1
oracle    66578      1  67339    4 TS   19 May21 ?        00:22:52 ora_lms2_anbob1
...

OK.

再看另一个问题环境Oracle 19.4 2-nodes RAC on RHEL 7.5

RAC Statistics
BeginEnd
Number of Instances:22
Number of LMS’s:4040
Number of realtime LMS’s:00(0 priority changes)

SQL> select * from v$bgprocess where name like 'LMS%';
PADDR              PSERIAL# NAME  DESCRIPTION                      PRIORITY     CON_ID
---------------- ---------- ----- -------------------------------- -------- ----------
0000001E01B628A0          1 LMS0  global cache service process     TS                0
0000001E01B65360          1 LMS7  global cache service process     TS                0
0000001E01B67E20          1 LMSE  global cache service process     TS                0
0000001E01B6A8E0          1 LMSL  global cache service process     TS                0
0000001E01B6D3A0          1 LMSS  global cache service process     TS                0
0000001E01B6FE60          1 LMSZ  global cache service process     TS                0
0000001E21AC8498          1 LMS3  global cache service process     TS                0
0000001E21ACAF58          1 LMSA  global cache service process     TS                0
0000001E21ACDA18          1 LMSH  global cache service process     TS                0
0000001E21AD04D8          1 LMSO  global cache service process     TS                0
0000001E21AD2F98          1 LMSV  global cache service process     TS                0
0000001E41A66B58          1 LMS6  global cache service process     TS                0
...

db alert log

2021-06-03T10:50:19.500768+08:00
LMON started with pid=22, OS id=98747
Starting background process LMD0
2021-06-03T10:50:19.527437+08:00
LMD0 started with pid=23, OS id=98749
Starting background process LMD1
2021-06-03T10:50:19.528918+08:00
* Load Monitor used for high load check
* New Low - High Load Threshold Range = [230400 - 307200]
2021-06-03T10:50:19.703222+08:00
Errors in file /u01/oracle/diag/rdbms/anbob1/anbob11/trace/anbob11_lms0_98751_98758.trc  (incident=873064):
ORA-00800: soft external error, arguments: [Set Priority Failed], [LMS0], [Check traces and OS configuration], [Check Oracle document and MOS notes], []
Incident details in: /u01/oracle/diag/rdbms/anbob1/anbob11/incident/incdir_873064/anbob11_lms0_98751_98758_i873064.trc
2021-06-03T10:50:19.711460+08:00
Error attempting to elevate LMS0's priority: no further priority changes will be attempted for this process
LMS0 started with pid=24, OS id=98751_98758
2021-06-03T10:50:19.800751+08:00
Errors in file /u01/oracle/diag/rdbms/anbob1/anbob11/trace/anbob11_lmsd_98808_98825.trc  (incident=873065):
ORA-00800: soft external error, arguments: [Set Priority Failed], [LMSD], [Check traces and OS configuration], [Check Oracle document and MOS notes], []
2021-06-03T10:50:19.815049+08:00
Error attempting to elevate LMSD's priority: no further priority changes will be attempted for this process
LMSD started with pid=50, OS id=98808_98825
2021-06-03T10:50:19.924836+08:00
LMD1 started with pid=104, OS id=98950
2021-06-03T10:50:19.924929+08:00
Starting background process LMD2
2021-06-03T10:50:19.944617+08:00
Errors in file /u01/oracle/diag/rdbms/anbob1/anbob11/trace/anbob11_lmsb_98797_98815.trc  (incident=873066):
ORA-00800: soft external error, arguments: [Set Priority Failed], [LMSB], [Check traces and OS configuration], [Check Oracle document and MOS notes], []
2021-06-03T10:50:19.945838+08:00
Error attempting to elevate LMSB's priority: no further priority changes will be attempted for this process
Starting background process LMD3
2021-06-03T10:50:19.949748+08:00

Note:
这套环境的LMS进程运行在TS模式，是因为在实例启动时遇到了ORA-800错误[Set Priority Failed]失败了。

检查oradism

oracle@anbob1a:/home/oracle/scripts_oracle$ ls -l $ORACLE_HOME/bin/oradism
-rwxr-x--- 1 oracle oinstall 147848 Apr 17 2019 /u01/oracle/product/bin/oradism

对于这个环境的owner和权限都是错的，修正后重启实例就可以解决。

也可以root用户使用chrt在线修改进程为RR mode。

# chrt -r -p 1 [lms pid]