Re: Did you once had to back out from an CPU implementation?

I have a real horror story. There was a CPU released in early 2005 for 8.1.7 on HP-UX.
We applied it. Then one day I needed to shut down the database and noticed that AFTER database was closed and unmounted it produced ORA-00600 error. Then the instance was shut down. I restarted the database without exiting SQL*Plus. You wouldn't believe what happened next. It started but mounted only one control file and brought online only one rollback segment. We didn't notice till next day when rollback tablespace became full. We investigated using "truss" to prove that it reads init.ora correctly, with all 3 entries for control files and all 20 entries for rollback segments. It did. We tested from Server Manager (svrmgrl) - same result.
Eventually Oracle admitted that it was a bug introduced in this CPU. It has a very simple workaround: exit SQL*Plus between shutdown and restart. Still it caused a few unplanned outages as people forgot to exit and we didn't want to run the database
with only one control file mounted.

On Apr 17, 1:36 am, Helma <helma.vi..._at_hotmail.com> wrote:
> Hi all,
>
> More a question of curiosity, but did you ever had to undo a CPU
> implementation? I've run across shops who
> a) never applied CPU's
> b) applied them without testing and
> c) read  the documentation to check if the cpu is relevant to them and
> any side-effects before installing.
>
> Now working at a 'c)' company, what are your experiences in this
> field ?
>
> Helma
> (secretly hoping to hear horrorstories)  ;)
Received on Fri Apr 18 2008 - 05:51:02 CDT