RE: linux kernel parameters

From: <dimensional.dba_at_comcast.net>
Date: Fri, 12 Feb 2021 20:53:09 -0800
Message-ID: <217d01d701c4$208601f0$619205d0$_at_comcast.net>





What was in the core file or trace files you uploaded for the SR. They may tell you why oracle is recommending increasing the values.

This could be a bug with rman that requires more semaphores than normal and this is their work around until they create a patch.    

From: oracle-l-bounce_at_freelists.org <oracle-l-bounce_at_freelists.org> On Behalf Of Mladen Gogala
Sent: Friday, February 12, 2021 7:03 PM
To: oracle-l_at_freelists.org
Subject: Re: linux kernel parameters  

Hi Jeff,

The valies of kernel.sem represent SEMMSL, SEMMNS, SEMOPM, and SEMMNI. SEMMSL is maximum number of semaphores in a semaphore set. SEMMNI is the maximum number of semaphore sets. SEMMNS is the maximum number of semaphores on the system. SEMOPM is the maximum number for semaphore operations per semop call. SEMMSL*SEMMNI should be equal to SEMMNS. Oracle's default settings do not make too much sense:

# oracle-database-preinstall-19c setting for kernel.sem is '250 32000 100 128'
kernel.sem = 250 32000 100 128

With the default settings, Oracle will consume 2 semaphore sets:

-bash-4.2$ ipcs -s

That is for 70 oracle processes:

-bash-4.2$ ps -ef|grep ora_|grep -v grep|wc -l
70

However, I usually change the number of semaphores like this:

kernel.sem = 64 65536 1024 128

That means that I have 65536 semaphores distributed into 1024 semaphore sets, each set containing 64 semaphores. When I check my semaphore usage now, it looks like this:

Those are 6 sets of 64 semaphores. There are 68 oracle processes on the system

-bash-4.2$ ps -ef|grep ora_|grep -v grep|wc -l
68

That means that we have approximately 6 semaphores per process:

scale=2
6*64/68
5.64

Now, let's make few connections and see what happens:

The number of consumed semaphores remains the same:

-bash-4.2$ ipcs -s

  • Semaphore Arrays -------- key semid owner perms nsems 0x8d2a3534 11 oracle 600 64 0x8d2a3535 12 oracle 600 64 0x8d2a3536 13 oracle 600 64 0x8d2a3537 14 oracle 600 64 0x8d2a3538 15 oracle 600 64 0x8d2a3539 16 oracle 600 64

That means that the semaphores are allocated by the instance, based on the processes parameter and not based on the number of running processes. The processes parameter is:

SQL> show parameter processes

NAME                      TYPE    VALUE 

------------------------- ------- -----
aq_tm_processes integer 1 db_writer_processes integer 1 gcs_server_processes integer 0 global_txn_processes integer 1 job_queue_processes integer 80 log_archive_max_processes integer 4 processes integer 320

The maximum number of processes is 320. On the other hand, 6 sets of 64 semaphores each means that we have 384 semaphores allocated for 320 processes. That means that the old rule of "1 semaphore per process + 50 to spare" still applies. However, with the default setting you allow 32000 semaphores on the system but you only allow 100 sets of 250 semaphores each. That means that you cannot utilize more than 25000 semaphores. Modifying the parameter according to the recommendation makes no sense whatsoever. BTW, I prefer more smaller sets, to have better granularity. My version is almost the same as yours

[mgogala_at_umajor ~]$ sql scott/tiger_at_ora19c

SQLcl: Release 20.4 Production on Fri Feb 12 21:48:37 2021

Copyright (c) 1982, 2021, Oracle. All rights reserved.

ERROR:
ORA-28002: the password will expire within 7 days

Connected to:
Oracle Database 19c Enterprise Edition Release 19.0.0.0.0 - Production Version 19.10.0.0.0

And yes, I did renew the password and alter password_life_time to unlimited. BTW, I was lazy as well. With 64 semaphores in the set, it doesn't make sense to allow 128 simultaneous operations per semop call. It is only possible to do a semop call on a single semaphore set, with one operation per semaphore. So, my recommended sessting is: kernel.sem = 64 65536 1024 64

On 2/12/21 8:20 AM, Beckstrom, Jeffrey wrote:

We have been pursuing a RMAN segmentation fauit issue with Oracle when doing a "recover plluggable database" command in 19.7. Oracle is suggesting the following change:  

According to the sysctl.conf that you uploaded, the following is the current KERNEL.SEM setting, as shown here.

kernel.sem = 250 32000 100 128

To avoid the segmentation fault, modify the /etc/sysctl.conf with the following:

kernel.sem=3000 32000 3000 128  

Have the kernel parameters changed with 19.7? Everything I Googled implies our values should be sufficient.    

Jeffrey Beckstrom

Greater Cleveland Regional Transit Authority

1240 W. 6th Street

Cleveland, Ohio 44113  

--

Mladen Gogala
Database Consultant
https://dbwhisperer.wordpress.com



--

http://www.freelists.org/webpage/oracle-l

Received on Sat Feb 13 2021 - 05:53:09 CET

Original text of this message