ipcs shows deleted shared segments, Oracle 9.2.0.4, HP-UX 11.23

[Andrei Lica_1]

On HP-UX 11.23 ( with March 2004 patches ) with Oracle9.2.0.4 after a clean DB shutdown some shared mem marked deleted but left allocated,
Example :
$ ipcs -m -b -p
IPC status from /dev/kmem as of Wed Feb 9 15:24:14 2005
T ID KEY MODE OWNER GROUP SEGSZ CPID LPID
Shared Memory:
m 0 0x411c0534 --rw-rw-rw- root root 348 556 556
m 1 0x4e0c0002 --rw-rw-rw- root root 61760 556 556
m 2 0x41200b6d --rw-rw-rw- root root 8192 556 568
m 90115 0x0c6629c9 --rw-r----- root root 17905056 1278 23990
m 4 0x06347849 --rw-rw-rw- root root 77384 1278 1322
m 8197 0x49145006 --rw-r--r-- root root 22908 1274 1322
m 98310 0x5e180024 --rw------- root root 512 1425 1425
m 671751 0x00000000 D-rw-r----- oracle oinstall 656367616 4812 4827
m 73736 0x00000000 D-rw-r----- oracle oinstall 555704320 17378 28749
m 57353 0xc4f6ea5c --rw-r----- oracle oinstall 757030912 19457 26511
m 32778 0xee7eabfc --rw-r----- oracle oinstall 656367616 4827 26645
m 1540107 0x444d568c --rw-r----- oracle oinstall 555704320 28749 25490
m 8204 0xb5721248 --rw-r----- oracle oinstall 555704320 23698 23698

It's enough to start the instance ( no application was started, only the db ) then stop it and some segments are left allocated ( with D flag )
If you exit from sqlplus session ( that issued shutdown command ) those segments disapear but not always.
CPID/LPID indicates non existent processes.

I've found this about my problem :
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=611062
but is not enough.
Does anyone encounter this problem ?
Is there an Oracle or HP-UX patch ?
HP guys don't want to help us if we don't proced applying all current HP-UX patches.

[Steven E. Protter]

Those segements should be released by oracle. The release will not be immediate, but it should happen.

If it does not happen, investigate two areas:

Memory leaks for your oracle version at metalink.

shared memory patches at itrc.hp.com

SEP

[Fred Ruffet]

9.2.0.4 is the one with the most bugs I know. Look at the list of patches for this version on metalink. *Maybe* a switch to 9.2.0.5 would be good for you.

Regards,

Fred

[Andrei Lica_1]

I know patch-ing oracle and hp-ux to the latest maybe a good idea.
But our Oracle admins app. consultants say oracle9.2.0.5 does not work ( or isn't certified ) with their application.
I/we want to make sure first that this can be solved by patching befor proceding to patch oracle/hp-ux. It's a production env.
Andrei

[TwoProc]

Andrei - what's the app ? B/c clearly this should upgrade should be covered in their certified path. I'd recommend 9.2.0.6 - but, if you go with 9.2.0.5 - watch out for the tool that automatically manages your PGA space - seems to be broken in 9.2.0.5 resulting in about a 1/2 speed database server unless you turn it off.

Down to a recommendation - well, you can get out of this easily by writing a script to delete the shared memory areas yourself. You can write a script to delete by ID or by Key using ipcrm - *IF* this is the only Oracle database on the box. Just make a script to find them, and delete them on shutdown, and make it part of your shutdown ritual.

[Andrew Cowan]

I seem to remember that this is a known problem with Oracle, and at one time they produced a free tool called "ipcclean". Maybe this could help?

[Andrei Lica_1]

John : iprm cannot remove them, That's the problem.
It's about a banking app. called flexcube

Andrew : I'll try that tool, hope it can remove those segments otherwise the only solution I know is reboot.

[Andrew Cowan]

The only other solution I have found for dropping hung connections etc, is to restart daemons such as inetd, and this can sometimes reset tables without a reboot.

[Fred Ruffet]

This is maybe a little bit off-topic, but John, could you be more specific about this problem on PGA auto managment bug ? Any metalink note ?

Regards,

Fred

[TwoProc]

Fred,

pga_aggregate_target=XXXXXXXX (some number)
workarea_size_policy=auto

I've found that while these worked just fine in 9.2.0.4 - but when I upgraded to to 9.2.0.5 I needed to turn them off as performance had gone down to fully 1/2 of what the system could normally handle at 100% in performance testing. I called Oracle Support at the time with no results, so I just did a "divide and conquer" on all cost-optimizer type settings until I isolated these two. By turning these off I regained full system performance.

There may have been bug fixes, etc for this since then, but I'm running fine without.