Re: RAC: Testing unplanned outage (1 instance goes down)

From: Karl Arao <karlarao_at_gmail.com>
Date: Mon, 9 May 2016 11:44:57 -0400
Message-ID: <CACNsJnfBYX--snFbOpC_SFnWHF1CCKVEtNn-NnfctF_w9b86gw_at_mail.gmail.com>



I have played around with expanding and reducing a node every minute and scheduled it in crontab
https://github.com/karlarao/OracleScheduledNodeAllocationTAF/blob/master/crontab_entry.md

the scripts are here
https://github.com/karlarao/OracleScheduledNodeAllocationTAF

then I would randomly kill the 2nd instance on the node 2 by executing

kill -9 `ps -ef | grep -i "ora_pmon_oltp2" | grep -v grep | awk '{print $2}'`

then while I'm doing the expand/reduce I have a DW and OLTP type SQLs running
https://github.com/karlarao/OracleScheduledNodeAllocationTAF/blob/master/run_dw.sh https://github.com/karlarao/OracleScheduledNodeAllocationTAF/blob/master/run_oltp.sh

then I would do the monitoring with the following scripts

sh mon_services.sh
sh mon_session.sh
sh px.sh

On this test I did not monitor the SCAN, that's good info BTW

-Karl

On Mon, May 9, 2016 at 10:30 AM, Igor Racic <igor.racic_at_gmail.com> wrote:

> Hi,
>
> When you do instance shutdown in sqlplus using shutdown (even abort)
> or alter pluggable database ... close, srvctl stop database or
> instance,
> Oracle gets that it's you stopping it and it treats it as planned outage.
> Killing pmon (without -9) works (likely as OS shutdown/restart).
> On the other hand, for application FCF tests, I am wondering what
> would be the lighest/fastest way to convince Oracle that unplanned
> outage just happened (and switch back to retry the test again).
>
> I have crossed somewhere that playing with OS network interfaces can
> do it, didn't test it yet.
> Any other candidate ?
>
> Thank you and regards,
> Igor
> --
> http://www.freelists.org/webpage/oracle-l
>
>
>

-- 
Karl Arao
Blog: karlarao.wordpress.com
Wiki: karlarao.tiddlyspot.com
Twitter: _at_karlarao <http://twitter.com/karlarao>

--
http://www.freelists.org/webpage/oracle-l
Received on Mon May 09 2016 - 17:44:57 CEST

Original text of this message