Re: zombies

Chuck Hamilton <chuckh_at_ix.netcom.com> wrote:
>jfruetel_at_ainet.com (John Fruetel) writes:
>>
>>You might want to see if you can fiddle with the TCP/IP KEEPALIVE
>>option on your host. I don't know about Solaris, but this can be
>>changed on SCO Unix from the default of two hours to something else (5
>>minutes in our case) and any "zombies" get nuked after 5 minutes.
>
>Wouldn't that also "nuke" processes that were running long ( > 5
>minute) queries?

No because the processes will still be there to talk to the backend. It doen't matter that they are busy from a Application POV because the SYSTEM side will take care of answering the Keepalive.

Also please note that this will do nothing to stop Orphan process that are still running something. ie) user selects something from a REALLY long table and reboots before it is finished.

In this case your keepalive timer will close the socket but the oracle process will not look at the socket until it is finished running. Not a big deal for most selects but on an endless pl/sql loop???

The onnly way to catch the above is to upgrade to SqlNet V2.? (not all versions of V2 have the appl timer feature although most new versions do) or to look for oracle processes that dont have a socket anymore and kill them manually.

BTW- My Unix/TCP bias is showing. I have no clue how a NT server with Pc frontends would solve this problem under IPX or if they would even have the above problem. Anybody know?

-- 
Regards, 

Lee E. Parsons                  		
Systems Oracle DBA	 			lparsons_at_world.std.com