Hi

My Fedora box has oracle10g installed freezez randomly. When it happen I cann't do anything, it just Hang. I need to reboot and everything will work fine again. What t can be?

I go to check in /var/log/messages and notice that after boot process it has something like

May 9 11:01:01 ora10g crond(pam_unix)[3509]: session opened for user root by (uid=0)
May 9 11:01:01 ora10g crond(pam_unix)[3509]: session closed for user root
May 9 12:01:01 ora10g crond(pam_unix)[3543]: session opened for user root by (uid=0)
May 9 12:01:01 ora10g crond(pam_unix)[3543]: session closed for user root
May 9 13:01:01 ora10g crond(pam_unix)[3587]: session opened for user root by (uid=0)
May 9 13:01:01 ora10g crond(pam_unix)[3587]: session closed for user root
May 9 14:01:01 ora10g crond(pam_unix)[3665]: session opened for user root by (uid=0)
May 9 14:01:01 ora10g crond(pam_unix)[3665]: session closed for user root
May 9 15:01:01 ora10g crond(pam_unix)[3725]: session opened for user root by (uid=0)
May 9 15:01:01 ora10g crond(pam_unix)[3725]: session closed for user root

until it freeze and i will reboot it again. What's this message mean? it look like some cron job open and close session every hour... not sure

So, any suggestion please, anylog, anywhere I should check

Thanks
poj

You only posted the file from /var/log/messages.
Did you check the oracle logfile for clues? (alert_<SID>.log)

This file is normally located in: $ORACLE_HOME/admin/<SID>/bdump

Another question: Are u using large memeory settings (like over 4 Gb)
Ik know there is a problem with Oracle 9ir2 on Redhat AS 3 with regards to this.

maybe u can check the parameter kernel.shmmax in the file /etc/sysctl.conf
This is the maximum amount of shared memory allocated on the server

thanks spaniel

I reset my box at about 05.56.35 on May 9 and this is log from alert_bcom.log at that time

...
Sun May 8 17:00:33 2005
Private_strands 7 at log switch
Thread 1 advanced to log sequence 581
Current log# 1 seq# 581 mem# 0: /u01/app/oracle/oradata/bcom/redo01.log
Sun May 8 22:00:43 2005
Private_strands 7 at log switch
Thread 1 advanced to log sequence 582
Current log# 2 seq# 582 mem# 0: /u01/app/oracle/oradata/bcom/redo02.log
Mon May 9 02:00:28 2005
Private_strands 7 at log switch
Thread 1 advanced to log sequence 583
Current log# 3 seq# 583 mem# 0: /u01/app/oracle/oradata/bcom/redo03.log
Mon May 9 05:59:01 2005
Starting ORACLE instance (normal)
LICENSE_MAX_SESSION = 0
LICENSE_SESSIONS_WARNING = 0
Picked latch-free SCN scheme 2
KCCDEBUG_LEVEL = 0
Using LOG_ARCHIVE_DEST_10 parameter default value as USE_DB_RECOVERY_FILE_DEST
Autotune of undo retention is turned on.
Dynamic strands is set to TRUE
Running with 1 shared and 18 private strand(s). Zero-copy redo is FALSE
IMODE=BR
ILAT =18
LICENSE_MAX_USERS = 0
SYS auditing is disabled
Starting up ORACLE RDBMS Version: 10.1.0.3.0.
System parameters with non-default values:
...

It look like nothing wrong just rotate log redo1->2->3

Parameters in sysctl.conf are

kernel.shmall = 2097152
kernel.shmmax = 2147483648
kernel.shmmni = 4096
kernel.sem = 250 32000 100 128
fs.file-max = 65536
net.ipv4.ip_local_port_range = 1024 65000

I have 1 Gb memory in that box. I read from somewhere that kernel.shmmax should set to half of physical memory(not sure). Should I set it to 5xxxxxxxxx??? or what it should be???

Looks to me like the shared memory you configured is 2 Gb (2147483648/1024/1024/1024 = 2),
You only have 1 Gb. I would set it to: 536870912 ( which is 512 * 1024 * 1024, which is half your physical memory)

On all boxes i have configured ( Oracle 8i and 9ir2), i did not need the rest of the parameters.
When i created the kickstart configurations for it i checked with the kernel sources (some .h file i don't remember ) and a lot, if not all of the parameters are already set to the values specified or higher and did not need changing.

And yes, Oracle recommendations are as you set them, so this is the way to set these parameters. But my guess is that you have to have enough memory. And another yes. I configured it on Redhat AS3 (enterprise stuff of redhat) which is of a different level of kernel and glibc.

The problem is that i dont have your init<SID>.ora file to see what you configured in the startup parameters of the oracle instance. If might be that the 512 Mb are not enough to start oracle.

You will find out if you:
- change the shmmax parameter
- stop the oracle instance
- invoke the command sysctl -p (as user root)
- restart the oracle instance and whatch the alert file (alert_bcom.log)

If it does not start, well then change it back

A few questions i do have:
At what size did you set your swap space?
Did you "yum update" the machine (if it is production make a backup first!!)?
Are you doing anything on the machine when it hangs (big selects, inserts, updates,etc,etc)?
You installed oracle without any errors? ( look in /tmp for some installation logging)
Did you install all of the oracle patches for 10g (look at metalink)?

My box has only 1 Gb. May be this is a problem???. I just see in document that it's recommend 2 Gb swap.
Never do "yum update" on that machine. I worry about oracle will not like new version gcc or sth else.
not sure i'm not use this box, I will find out for this issue
no error log in /tmp and never patch

now I'm wait for the time(when it's down again or when low traffic) I will try new kernel parameter "kernel.shmmax =536870912". If it still not stable may be I'll reinstall everything and make it 2Gb swap.

I would recommend yum updating the machine.

To be able to recover from that some hints might be usefull , so below some stuff i figured out when creating my oracle installation scripts:
-----------------------------------------------------------------------------------------------------------------

For Oracle you have to install compat rpm's
I installed the following rpm's above the core redhat installation to be able to install oracle:

glibc-kernheaders-2.4-8.34.i386.rpm
glibc-headers-2.3.2-95.20.i386.rpm
glibc-devel-2.3.2-95.20.i386.rpm
compat-gcc-7.3-2.96.128.i386.rpm
compat-libstdc++-7.3-2.96.128.i386.rpm
compat-libstdc++-devel-7.3-2.96.128.i386.rpm
compat-gcc-c++-7.3-2.96.128.i386.rpm
XFree86-xauth-4.3.0-62.EL.i386.rpm (this is to be able to use the installer)

(chech your machine which compat libraries are installed ( they are probably of another version then mine, but it should basically be the same libraries)

After that i:
- moved /usr/bin/gcc to /usr/bin/gcc.backup
- moved /usr/bin/g++ to /usr/bin/g++.backup
- renamed /usr/bin/gcc296 (which is from the compat libraries) to /usr/bin/gcc
- renamed /usr/bin/g++296 to /usr/bin/g++

i put the following stuff in the .bash_profile of user oracle:

# Oracle specific
export ORACLE_BASE=/opt/app/oracle/9.2.0
export LD_ASSUME_KERNEL=2.4.19
export ORACLE_HOME=/opt/app/oracle/9.2.0

After this i installed oracle patch 3006854. It compiles some library (/etc/libcwait.so)
and puts it in your /etc/ld.so/preload. I know there is a simular patch for 8
and i am not sure, but i think i have seen one for 10g. Get it at oracle metalink.
If you dont have a account at metalink(www.oracle.com), then get one!.

This allowed me to run the oracle installer and install without errors and have about 8 oracle machines running without any pain for more then a year now

If you have the above stuff to check/reconfigure after the update it should allow you to do a yum update without any real danger, But as they say, warranty until the frontdoor. Be carefull and don't blame me if your up all night to restore the machine

I dont think twice the memory as swap space is the problem (or u should have very high loads) if your machine/database can function with the 512 Mb Shared memory it will probably never swap. Normally out of memory messages are shown on the console.

One other remark/question i have:
Do you have the fedora box running in runlevel 3 (basic ascii) or runlevel 5(graphical X windows). A collegue of mine once encountered a problem with a redhat box which ran in runlevel 5 (x-windows) and was unstable too. He put the machine back in runlevel 3 and everything was stable again. He did not investigate further
If you need graphics for management then use another PC with X running and start the management stuff over the network.

Oracle is not officially support FC3 so I'm not sure about compat rpms. But I'm update some recommended rpm and set kernel parameter following this site.

http://www.oracle-base.com/articles/...nOnFedora3.php

and yes my box is in run level 3

may be I need to fight with my boss for official support from oracle or for RHES

What you did (as described on the website) is basically what i described (you installed some compat rpms, but of a higher version), and with some tweaking to convince the installer that het is not installing on redhat but on RHES3.

And yeah, convince your boss.

To be able to install oracle on a Redhat box i have found it to be very "handy" to have to officially support OS and a support contract at Oracle.
Without those you are missing some important patches (oracle in this case) to be able to install the boxes on linux and keep it stable.

But, did you try the shmmax setting i proposed already? It might just solve the problem and make the rest of the discussion obsolete.