Re: Killed? Mutating serial Number

kasten_at_brookings.net (Chris Kasten) writes:

>I recently have seen reference to the following scenario, but did
>not pay close attention...
 

>- A session is doing massive updates to the database.
>- That session is killed (say by user's request).
> (Alter system kill session 'x,y'; )
>- Session is marked 'KILLED' in v$session.
>- After rollback is complete it goes away.
 

>That seems to sum up the normal process, right?
 

>Sometimes, however, the serial number begins to increment
>itself and appears to never rollback or go away (even tho
>it is statused as 'KILLED'). We eventually end up having
>to bounce the instance.
 

>The questions:
>1) Why does that happen?

This was a known bug in Oracle 7.1.3. Oracle support told me that it would be fixed in the next release. I guess not.

>2) How can we clean those rogue sessions up without restarting
> the DB.

You mention that you have to stop and restart the instance (which works). I have waited patiently and the runaway session eventually goes away. It also appeared that issuing:

alter system checkpoint;

speeded up the process a little.

>Running 7.2.2.4 Enterprise on WinNT (with parallel and replication).
>I've seen this happen in other versions/platforms.

>-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
>Chris Kasten
>Programmer/Analyst
 

>#include <std/disclaimer.h>
 

>"File not Found:
> (A)bort, (R)etry, (C)all Geordi in Engineering?"
>=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=

-Ari Kaplan
email :akaplan_at_interaccess.com
homepage :http:/homepage.interaccess.com/~akaplan Received on Wed Feb 28 1996 - 00:00:00 CET