Re: Instance , what instance.

>> kill -9 pmon, dbwr, ckpt, smon etc, in order of PID.
>
>What did a 'ps -ef | grep ora' showed afterwards?

All the backgrounds really are dead, I made sure that the only Oracle things left alive were the shadows (or dedicated servers to the youngsters on the group) viz: oracle{SID}.

> Session 3 should
 be hanging.. right?
>For it to access (some?) virtual views seem to imply that these are either
 kept
>locally within that process (does not make sense), or that some shareable
 memory
>are still available and accessible.. but then who is doing the writing to
file

Session 3 is actually doing reads, not writes. The reason the 'instance' keeps going is that nothing has removed the shared memory segment (i.e. the SGA) where most of the virtual views reside. Any foreground process that needs LGWR or DBWR to do some writing hangs. Any process that is only reading continues, since foregrounds read into the SGA. The direct loads actually ran (I think) to the point where they tried to bounce the HWM, then hit the brick wall of a non-existent LGWR.

I think the visibility of this feature is partly due to the new 'mid-point insertion' algorithm - pre 8.0, it was possible for a read to force a dirty block to be written, and therefore reader processes would very soon start to hang in the circumstatnce I defined. In 8.1, reads don't flush dirty blocks to disk with any regularity (the dirty blocks tend to get bounced to the hot end of the buffer when they first hit the tail of the LRU), so it will generally take much longer for a read to be in a position where it has to wait for DBWR to write before it can continue.

On the minus side, it could cause confusion of the type: "database is down"/"can't start - instance is up"

On the plus side, when you eliminate the memory segment the database can restart are recover without harm.

> Who the hell really killed JFK!?

This is explained very clearly in Red Dwarf.

>PPS. Do you think it is too early for my 3rd cup of coffee?
Definitely - I restrict myself to one mug in the morning, and when that pint is empty, that's it until elevenses.

--
Jonathan Lewis

Seminars on getting the best out of Oracle
Last few places available for Sept 10th/11th
See http://www.jlcomp.demon.co.uk/seminar.html




Billy Verreynne wrote in message <9lag96$aqf$1_at_ctb-nnrp1.saix.net>...


>"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote

>

>> Here's an amusing little test you might like to try.

><snipped>

>

>> From operating system:

>>   kill -9    pmon, dbwr, ckpt, smon etc, in order of PID.

>

>What did a 'ps -ef | grep ora' showed afterwards?

>

>Twisted logic (is there anything else?) says that if the system processes

 are


>dead, the local ones will no longer work. Session 3 should be hanging..

 right?


>For it to access (some?) virtual views seem to imply that these are either

 kept


>locally within that process (does not make sense), or that some shareable

 memory


>are still available and accessible.. but then who is doing the writing to

 file


>1? Who is that local process talking to? Who the hell really killed JFK!?

>

>Er.. is there a detective in the house? ;-)

>

>--

>Billy

>PS. You've caused me to think and made my head hurt, damn it man!

>PPS. Do you think it is too early for my 3rd cup of coffee?

>

>