Re: Killed session

  • From: "Jared Still" <jkstill@xxxxxxxxx>
  • To: Mayen.Shah@xxxxxxxxxx
  • Date: Wed, 12 Nov 2008 16:53:42 -0800

On Wed, Nov 12, 2008 at 6:24 AM, <Mayen.Shah@xxxxxxxxxx> wrote:

>
> Oracle 9.2.0.8 on Solaris 9
>
> In one of my production database session was long running.  Session was
> killed last night. Since then session is marked as KILLED in v$session. This
> killed session is holding locks on few tables and causing further problems.
>
> USED_UBLK in v$transaction is constant at 287 since late last night.
>
> I identified unix process and killed at os level two hours ago, still
> session would not go away. (os process is gone already)
>
> What else should I be looking at? How do I get rid of KILLED session?
>

When I worked on Solaris (been a few years) we had the same problem
regularly.

One thing I found was that if you first kill the session in the database,
you're sunk.
It was necessary to first use kill -9 at the OS level to kill the process.

If we didn't do that, the only way out was to bounce the database.

This is referring only to sessions holding a lock of course.

Things may have changed between versions since the time I worked on
Solaris, but it sounds remarkably similar.

Actually, I had seen that happen earlier on DG/UX, back in 7.x days.


>
> I do not have luxury to recycle database.
>

Once it gets to that state, and you can verify that there is nothing being
rolled back by PMON, then there's likely nothing else to do.

Jared

Other related posts: