Regards,
Larry G. Elkins
<mailto:elkinsl@xxxxxxxxxxx> elkinsl@xxxxxxxxxxx
214.695.8605
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] On ;
Behalf Of Kellyn Pot'Vin-Gorman
Sent: Monday, April 25, 2016 12:40 PM
To: Jessica Mason <jessica.masson85@xxxxxxxxx>
Cc: ORACLE-L <oracle-l@xxxxxxxxxxxxx>
Subject: Re: Excessive Logical IOs against which Table/Index
Well, you're quite "up a creek with only half a paddle"! :(
I would definitely say, due to the workload type, (so many environments do have
monthly or monthend processing) a business case to have them bump it up to five
weeks so that your research could be more productive.
Second, it may be helpful to have this for reference:
https://docs.oracle.com/cd/B28359_01/server.111/b28274/instance_tune.htm#CACGBHBD
I'll let a couple others chime in and see what other recommendations folks have
to locate what changed, but v$sqlarea and v$io* objects sounds like a good
start...
Kellyn
<http://d3mod6n032mdiz.cloudfront.net/thumb2/d/b/a/dbakevlar/dbakevlar-105x70.jpg>
Kellyn Pot'Vin-Gorman
about.me/dbakevlar
On Mon, Apr 25, 2016 at 11:31 AM, Jessica Mason <jessica.masson85@xxxxxxxxx
<mailto:jessica.masson85@xxxxxxxxx> > wrote:
Hi Kellyn,
(1) Historical data is only stored for 21 days in the production database and
this is a monthly job. So, there was no optimal plan to compare ie the whole
exercise.
(2) The information was gathered from v$ views.
I have only SELECT_CATALOG_ROLE granted to me for this database and have to
troubleshoot all the issues using v$ views only.
Thanks
JM
On Mon, Apr 25, 2016 at 10:52 PM, Kellyn Pot'Vin-Gorman <dbakevlar@xxxxxxxxx
<mailto:dbakevlar@xxxxxxxxx> > wrote:
I'm going to take a shot at this with questions instead of answers, (sorry!)
1. Do you have the execution plans for the optimal vs. the plan that was used
during this execution?
2. Did you gather this information from simple V$views or did you trace the
session?
Thanks,
Kellyn
<http://d3mod6n032mdiz.cloudfront.net/thumb2/d/b/a/dbakevlar/dbakevlar-105x70.jpg>
Kellyn Pot'Vin-Gorman
about.me/dbakevlar
On Mon, Apr 25, 2016 at 11:10 AM, Jessica Mason <jessica.masson85@xxxxxxxxx
<mailto:jessica.masson85@xxxxxxxxx> > wrote:
Hello List,
Last week, I was involved in a production issue, where a data load job, which
normally takes few hours to complete, had been running for more than 48 hours.
I tried to take the following systematic approach to identify the cause -
Step 1 - Identify the session and started profiling it. All the time, the
session was on CPU.
Step 2 - To understand why the session was burning CPU, the v$sesstat view was
queried and below were the top statistics that were changing :
43126075162624 logical read bytes from cache
240440566773 table scan rows gotten
2632208820 session logical reads
2632206511 consistent gets
2632206511 consistent gets from cache
2632205708 consistent gets from cache (fastpath)
Step 3 - Next, I wanted to know the object ( table/index) against which these
logical IOs were happening so that I could focus on the operations, involving
these objects, in the execution plan but didn't know which view to query.
The above information could have saved us lot to time to identify the cause (
in this case, an unique index was dropped and Oracle was doing FTS on a table
which was referred 6 times in the query fetching million of records).
So, my questions to the list is that which v$ view should I have checked?
Or is there a better approach to troubleshoot such issues?
Thanks
JM
