RE: Extended database statistics
Date: Sat, 25 Feb 2017 14:27:54 -0500
Message-ID: <081001d28f9d$43fff110$cbffd330$_at_rsiz.com>
Turn everything adaptive off, and then drop them one last time?
This is unlikely to be the never found yet silver bullet, but it *might* be a good idea for EBS where they have a very particularly specific stats collection routine as mentioned by Tim much earlier, and only applying the specific extended stats that FND_STATS patched dictates or that you specifically research as beneficial to your specific data texture in EBS.
Because of the power and flexibility of the flexfield structures and the variability of charts of accounts and master item lists (as two examples), the general case optimization of EBS is probably intractable. A good starting point is what they have provided and then augmented by knowing your data "texture" and troubleshooting anything that seems to consume significantly more resources than it should consistently.
Particularly in EBS this is distinctly NOT usually the same as tuning the most expensive thing you have. Usually in EBS that is a mainline job part and parcel of the basic major functioning as general ledger, manufacturing, and so forth. Your mileage may vary, but often you'll find that while remaining expensive there is no fat on the bones of these processes.
Of course the biggest single thing remains that workflow and the accumulation of CCMGR job logs has never been purged. THAT is pure fat easily rendered.
mwf
-----Original Message-----
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Hameed, Amir
Sent: Friday, February 24, 2017 4:21 PM
To: Jonathan Lewis; alfredo.abate_at_gmail.com; mwf_at_rsiz.com
Cc: 'Maris Elsins'; tim.evdbt_at_gmail.com; oracle-l_at_freelists.org
Subject: RE: Extended database statistics
The statement in question belonged to the auto-invoice job. It statement was initially using INDEX_N6 and was taking over 30 minutes to complete. The row set operation from TKP showed that it was scanning over 155m rows from NDEX_N6. I forced it to use INDEX_N7 via a hint and it finished within a second. Because it was a standard EBS code, we went to Oracle and asked that we need this statement optimized. After doing further analysis, it was determined that extended statistics existed on the underlying table; there were 10 of them. After we dropped those extended statistics, the statement automatically started using INDEX_N7. So, the extended statistics were messing up the execution plan.
There are currently a total of over 4500 extended statistics on the standard tables. I was hoping if there was a way to either disable these extended statistics instead of dropping them but I am not sure if they can be disabled.
-----Original Message-----
From: Jonathan Lewis [mailto:jonathan_at_jlcomp.demon.co.uk]
Sent: Friday, February 24, 2017 4:01 PM
To: alfredo.abate_at_gmail.com; Hameed, Amir <Amir.Hameed_at_xerox.com>;
mwf_at_rsiz.com
Cc: 'Maris Elsins' <elmaris_at_gmail.com>; tim.evdbt_at_gmail.com;
oracle-l_at_freelists.org
Subject: Re: Extended database statistics
If you have the 10132 trace - or pull the plan from memory with the other_xml - there's information about the number (and nothing else) of sql plan directives viewed and used; but as soon as a directive is replaced by a column group information the directive is no longer counted.
Regards
Jonathan Lewis
From: oracle-l-bounce_at_freelists.org <oracle-l-bounce_at_freelists.org> on behalf of Mark W. Farnham <mwf_at_rsiz.com> Sent: 24 February 2017 20:55:09
To: alfredo.abate_at_gmail.com; Amir.Hameed_at_xerox.com Cc: 'Maris Elsins'; tim.evdbt_at_gmail.com; oracle-l_at_freelists.org Subject: RE: Extended database statistics
Regarding whether or not you can see from the plan whether extended stats helped you arrive at that plan, I'm not aware of anything short of a Wolfgang trace that shows you what the CBO process was in arriving at the plan.
It would probably be difficult to remove the extended stat columns and re-run the exact query again to see if you obtained a different plan without them.
And if you've got adaptive features activated, good luck telling WHY the plan changed anyway. I think it checks the race results at Las Vegas. (Not really. But right now there are more factors going on than I can keep track of and I'm not even sure which factors are taking place that we cannot see. If there is an event like the Wolfgang trace that can cause "adaptive" to spit out something when it makes a change and tells why that might be helpful. I'm not sure whether adaptive changes are reported in a Millsap trace. I have never seen them there, but not seeing evidence is not the same as not happening.)
mwf
From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Alfredo Abate
Sent: Friday, February 24, 2017 12:54 PM
To: Amir.Hameed_at_xerox.com
Cc: Maris Elsins; tim.evdbt_at_gmail.com; oracle-l_at_freelists.org
Subject: Re: Extended database statistics
Amir,
Both Tim and Maris have given you excellent advice already. The only other thing I will mention is to review MOS doc # 396009.1 to make sure you have the database initialization parameters set accordingly for EBS.
The parameter you mentioned, OPTIMIZER_ADAPTIVE_FEATURES, is one of these parameters on the list that needs to be set to FALSE.
Alfredo
On Fri, Feb 24, 2017 at 9:36 AM, Hameed, Amir
<Amir.Hameed_at_xerox.com<mailto:Amir.Hameed_at_xerox.com>> wrote:
We found thousands of extended statistics on standard tables. According to
Oracle, they are collected if OPTIMIZER_ADAPTIVE_FEATURES is set to TRUE and
this is what I want to validate.
Also, if there a way to find out from an explain plan if extended statistics
were used for the pan?
From: Maris Elsins [mailto:elmaris_at_gmail.com<mailto:elmaris_at_gmail.com>] Sent: Friday, February 24, 2017 10:23 AM To: tim.evdbt_at_gmail.com<mailto:tim.evdbt_at_gmail.com> Cc: Hameed, Amir <Amir.Hameed_at_xerox.com<mailto:Amir.Hameed_at_xerox.com>>; oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org>
Subject: Re: Extended database statistics
If I remember right, then the column sets that FND_STATS collect the extended stats for are defined in FND_EXTNSTATS_COLS (for eBS). If you have more extended stats than defined in this table, then there's something else collecting them too.
--- Maris Elsins _at_MarisElsins<https://twitter.com/MarisElsins> www.facebook.com/maris.elsins<https://www.facebook.com/maris.elsins> On Fri, Feb 24, 2017 at 5:16 PM, Tim GormanReceived on Sat Feb 25 2017 - 20:27:54 CET
<tim.evdbt_at_gmail.com<mailto:tim.evdbt_at_gmail.com>> wrote:
Amir, In releases of EBS prior to R12.2, patching to FND_STATS might be necessary to make access to extended statistics available, according to this support white paper<https://support.oracle.com/epmos/main/downloadattachmentprocessor?atta chid=1586374.1%3ADOC1586374_A&docType=WHITE%20PAPER&action=download>. If those patches have not been applied to FND_STATS, then you can eliminate EBS from your research. Hope this helps... -Tim On 2/24/17 08:05, Hameed, Amir wrote: Thanks Tim. The EBS version is 12.1.3. From: oracle-l-bounce_at_freelists.org<mailto:oracle-l-bounce_at_freelists.org> [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Tim Gorman Sent: Friday, February 24, 2017 10:02 AM To: oracle-l_at_freelists.org<mailto:oracle-l_at_freelists.org> Subject: Re: Extended database statistics Amir, The version of EBS is important to share as well, but since the underlying database is 12c, I'm guessing it likely that EBS is R12.2.x? If so, please be aware that the FND_STATS package in EBS R12.2 supports the gathering of extended statistics<https://docs.oracle.com/cd/E26401_01/doc.122/e22954/T202991T17430 6.htm>, and you may want to check the settings in use to determine if the presence of the extended statistics is caused by EBS settings, rather than something in the RDBMS parameters? Hope this helps... -Tim On 2/24/17 07:55, Hameed, Amir wrote: Hi, The Oracle database version is 12c (12.1.0.2). The application is Oracle E-Business Suite. While working on a performance issue, we discovered that there were a lot of extended statistics present in the database for standard EBS tables. We do not create extended statistics on any table as part of our statistics gathering process. The following parameters were set to TRUE: optimizer_adaptive_features optimizer_adaptive_reporting_only I am trying to understand the following: 1. Does Oracle collect extended statistics automatically if the optimizer_adaptive_features parameter is set to TRUE? 2. Is there a way to find out from the explain plan if extended statistics were used in the plan? Thank you, Amir -- http://www.freelists.org/webpage/oracle-l -- http://www.freelists.org/webpage/oracle-l