Re: Shutdown hangon.

Yup, this is true. It happens to me on my training courses -and I know damned well I'm the only one logged in to the demo database (barring the odd swine of a student!) , and I have to revert to an abort, too.

I'd almost buy the deadlock thing except that, occasionally, it happens when I've started up, and do an immediate Immediate shutdown (if you get my drift!), and there's not a transaction in site.

I was once told that when you issue a shutdown immediate, there's a conversation to be had with the Agent and Listener. Not sure if that's true or not, but maybe that's a source of potential delay.

Regards
HJR

--
--------------------------------------------------------------------------
Opinions expressed are my own, and not those of Oracle Corporation
Oracle DBA Resources:               http://www.geocities.com/howardjr2000
--------------------------------------------------------------------------

"Greg Stark" <greg-spare-1_at_mit.edu> wrote in message
news:878zr83zln.fsf_at_HSE-MTL-ppp62507.qc.sympatico.ca...


>

> "Howard J. Rogers" <howardjr_at_www.com> writes:

>

> > The delay in a shutdown immediate (I've seen one take 6 hours) is

 because,


> > having booted the users off, Oracle then rolls back whatever

 transactions


> > were pending at the time of the shutdown command being issued.  If you

 have


> > a bezillion pending transactions that thus need to be rolled back, you

 can


> > expect the shutdown not to happen for some time.

>

> FWIW We've had the same experience. We do a regularly use shutdown

 immediate


> to take a cold backups; it normally takes a few seconds to kill all the

> sessions and roll back all the transactions. Occasionally it simply hangs

 and


> appears to be ready to hang indefinitely until an admin comes along and

 uses


> shutdown abort. This seems to happen about 1% of the time or so.

>

> It should not be possible for it to be a user that has left an uncommitted

> transaction, since shutdown immediate should kick off any such users. It

> should not be possible for it to be an inordinate amount of rollback to

 apply


> since it's at the same time and under the same load consistently.

>

> I am convinced this is simply an Oracle bug. A dead lock of some kind in

 the


> shutdown code, or a resource leak that makes Oracle think some session is

> still active after it's been killed, or who knows. We've opened Oracle

 TARs


> but they never seem interested in actually finding such bugs since they're

> hard to reproduce.

>

> --

> greg