Re: CPU usage after killing an Oracle session

From: Jgreene <jgreene_at_aol.com>
Date: 1995/06/22
Message-ID: <3scskb$hnh_at_newsbf02.news.aol.com>#1/1


Reply to: joanne_at_eosdev1.gsfc.nasa.gov (Joanne Woytek)

> Has anyone experienced situations where an Oracle session is cancelled,
> either through the alter system kill session command or with a ctrl-C
 and
> Oracle continues to chug along using 90% of the CPU? This has happened
> to us a number of times now from different users under different
> circumstances. I plan on trying some systematic testing but was
> wondering if others had ideas as to what might be happening. We are
> running Oracle 7.1.4 on an SGI IRIX 5.3 system.

If you think that is fun, try watching 7 out of 12 processors on a Sequent being tied up with hanging processes. The problem on this system occured with client-server applications where the user hit control-C or turned off their machines. The problem comes in that SQL*Net version 1 is not very smart when it comes to monitoring its connections. Therefore, you have a SQL*Net session that continuously polls to see where its client is when there is no way for it to re-establish a connection. The solution proposed by Oracle was to use the latest version of SQL*Net version 2 with some timeout parameters set on as described in the SQL*Net literature. We would up warning the users to call the DBA when they did this so that we could kill the processes.

We killed the sessions within SQL*DBA and this killed the associated operating system processes. This does not match up with your problem when issuing an alter session command, however it could be a starting point. Question- can you (as the Oracle owner perhaps) kill these run-away processes at the operating system level? Hope this at least gives you something interesting to think about. Received on Thu Jun 22 1995 - 00:00:00 CEST

Original text of this message