Re: RAC gv$session

Re: RAC gv$session SID and SERIAL# are not guarenteed unique across instances. INST_ID is the last part of that implied primary key, for all GV$ views. Look at V$FIXED_VIEW_DEFINITION

The data for each instance, including memory-based V$ views like V$SESSION, is maintained solely within that instance. When necessary, those GV$ views are populated using cross-instance queries via parallel execution “slave” (i.e. “ora_pXXX_SID”) processes. Exceptions include the V$ views that are based in the control file, such as V$DATABASE, V$THREAD, V$DATAFILE, etc. whose data is visible to all instances through the shared control files. At least, that is how it was done for 8/8i OPS — I haven’t verified whether 9i RAC has found something more clever involving cache fusion and the “block-shipping” processes or something. The GV$ views did not exist in Oracle7 and prior — we had to write them custom. Can anyone confirm or correct about how 9iRAC displays GV$ info?

Still, although you can view SID and SERIAL# from across the cluster (though GV$ views), you can only utilize them in “DBMS_” package calls or from ALTER SYSTEM KILL SESSION, etc. from the appropriate instance...

on 3/11/04 7:48 AM, Kommareddy, Srinivas (MED, Wissen Infotech) at Srinivas.Kommareddy@med.ge.com wrote:

Hi All,

We have 2 nodes set for RAC.

When I try to kill a session from node 1, is the sid,serial# going to be unique among the nodes,
Or is there any possibility that another session on node 2 will be killed (if sid, seiral# are same for another session on node 2).

Does my select sid,serial# checks v$session or gv$session.

Can someone through some light on this.

Tx and Regards,
Srinivas

Re: RAC gv$session

Tim Gorman

2004-03-11