Re: Dead Cllient Connection Cleanup

TONY, GAST wrote:
>
> As most places that have PCs connecting to Oracle, we have users that
> ocaisionally submit long running queries. Thinking that their PC is
> hung up, instead that their query is bad, they shut down and/or power
> off their computers. This leaves the query still running. Are there
> any settings in the Oracle instance, or the UNIX server that can be set
> to help out with this? The Oracel server is running on SunOS 5.5.1

I am running into this very same problem on SunOS 5.5.1. I have tried the DCD feature and found it to be helpful but not a complete solution. DCD will send out a probe to the PC every X number of minutes. If the PC responds, it will keep the server process alive. It does not check the actual SQL*NET connection to the PC. You could have a a user reboot and be back into Windows before the DCD probe. In this case, DCD will not do anything.

In speaking with Oracle on this matter, they suggested that I set idle time restrictions in the default user profile. I did this and discovered a nasty Oracle bug in 7.3.2.3. If you are using parallel query and resource limits (idle time at least), the parallel query servers will eat all available CPU time even when Oracle reports them as being idle.

Even with the resource limit, the process will be marked as sniped. It will not be cleaned up until it can notify the user to this fact. Since the connection is actually gone anyway, this is no help either.

I have also tried to tie a specific oracle process back to a unix process. The idea here is to query the database for processes that have been sniped and kill them at the OS level. I have not been able to accomplish this. Oracle network support says that this can not be done with SQL*NET V2. Received on Wed May 21 1997 - 00:00:00 CDT