RE: alter system v shutdown abort

  • From: "Jesse, Rich" <Rich.Jesse@xxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Fri, 15 Sep 2006 08:48:35 -0500

[quickly checks ipcs -a]

Whew!  Three contiguous shared memory segments for the three Oracle 
instances...  Then again, I would imagine that Oracle wouldn't have let me 
restart the instance if the previous incarnation of it hadn't released it's 

I hadn't seen that one before, Joel!  Did you have a zombie process hanging 
onto that segment or what?  There's a method to tell what process(s) are 
attached to a given segment, but I haven't had to use it except for testing 
(that I can remember!).


-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On 
Behalf Of Joel Garry
Sent: Thursday, September 14, 2006 6:27 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: alter system v shutdown abort

Jesse, Rich wrote:

>I had one such occasion on under HP/UX 11.11 just last week
>a dev DB that gets rebuilt from scratch weekly.  The sniped process
>prevented the release of shared memory even after a SHUTDOWN ABORT
>(confirmed by ipcs -a), which prevented the new instance from starting
>up.  I haven't needed to take time to troubleshoot it, since it's just
>dev DB.

I've also seen this on 9206 on hp-ux 11.11.  Even ipcrm couldn't remove
it, I had to bounce the production box to release the shared memory.
Rare, but it can happen.  

If all you (Queen Roo-Roo) are doing is killing off users for a backup
(and you should investigate if that is really necessary for the backup,
or just a means of preventing undo problems because people don't log off
correctly), you can script OS process kills with something like:

Echo "Killing list:"
ps -ef|grep oracle${ORACLE_SID}|grep -v grep|grep -v <any other things
you don't want to kill>|awk '{print "kill -9 "$2}' >>
<then the same line again without the redirect to output what is going
to be killed>
echo "[End of list]"
if [ -f $$kill_oracle_attaches ]
then chmod +x $$kill_oracle_attaches
rm $$*

This is handled in oracle by PMON, while the alter system is handled by
SMON, who may never get around to it.  Sessions killed with alter system
not releasing locks is pretty common.

Joel Garry 


Other related posts: