Re: Re: Fixing Performance issue with less selective columns
Date: Thu, 26 Aug 2021 10:00:28 +0200 (CEST)
Message-ID: <261226719.8338.1629964828902_at_bluewin.ch>
Hi Lok,
I have to say that I lost track of which columns are how selective.
So what I would do is the classical bitmap use, which are single column Bitmap which can be combined.
Maybe nót the best way, but the most flexible.
For all column where <> and not null is used, you should create a bitmap index.
Thus we have one bitmap on NVL (I.MA_FLG, 'N') and one on NVL (I.D_UNMTCH, 'N'). For not <> only bitmap would work.
For is null and is not null we can also make a single column bitmap index.
It all depends how often this table is inserted and how often the columns are updated. How much is too much is hard to decide and quite often the truth will show only at runtime. There is some risk on bimap in an OLTP, always.
For all the other columns you can create a combined B*Tree index.
There is an other option like creating a big combined B*Tree with the columns compare with = in the leading positions.
You can use the other columns as filter . This is a low risk option
E.g. something like this (in that sequence) TRIM(I.M_TXT) , PT_Code, DC_CODE, D_CUR_CODE, ED_AMT, PR_CTGRY, NVL (I.MA_FLG, 'N'), NVL (I.D_UNMTCH, 'N'), WOF_DATE, PE, PT_MCODE.
How good this index is will show in the combined selectivity of the = compared columns.
You can check this by :
select count(*) from (
select distinct TRIM(I.M_TXT) , PT_Code, DC_CODE, D_CUR_CODE, ED_AMT, PR_CTGRY
from PP_IN_TAB)
;
I the selectivity is not good enough, you can increase the selektivity by using a transformation like Mark described it.
Regards
Lothar
----Ursprüngliche Nachricht----
Von : loknath.73_at_gmail.com
Datum : 25/08/2021 - 21:33 (MS)
An : l.flatz_at_bluewin.ch
Cc : oracle-l_at_freelists.org
Betreff : Re: Fixing Performance issue with less selective columns
Thank you Lothar. Actually this is an OLTP kind of database and I do see UPDATE queries on these table columns. Need to see in detail about the frequency of those and how concurrent they are. I am Not very familiar with the usage of bitmap indexes , so trying to understand what exact column you are suggesting for the bitmap index to help this query?
On Wed, Aug 25, 2021 at 11:45 PM Lothar Flatz < l.flatz_at_bluewin.ch> wrote:
This scenario cries for bitmap indexes.
Bitmap Indexes can deal with "not equal" as well as "is null".
The columns seems to be low cardinality too.
The only open question is how often these columns get updated. (
https://asktom.oracle.com/pls/apex/asktom.search?tag=bitmap-indexes-and-locking).
Regards
Lothar
Am 25.08.2021 um 19:19 schrieb Mark W. Farnham:
unfortunately you keep nearly all the rows of both MA_FLG and D_UNMTCH, so this query is the opposite of those indexes being useful.
IF you were looking for ‘Y’ instead of not ‘Y’ on either one it would be extremely good. I didn’t see initially that these two columns are extremely inclusive.
I think Sayan was checking that in his query request. MA_FLG could reject at most about 6 million rows, so that’s pretty worthless.
From: Lok P [mailto:loknath.73_at_gmail.com]
Sent: Wednesday, August 25, 2021 1:09 PM
To: Mark W. Farnham
Cc: Sayan Malakshinov; Oracle L
Subject: Re: Fixing Performance issue with less selective columns
Thank You Mark. I may be wrong but in this situation I was unable to think of any other way we could make this query faster , so I was thinking of creating a new index. If there exists any other way to make this query faster without creating any new index that would really be helpful. I am not able to get your point fully, If you can help me understand it a bit more here please. Below is the data pattern for MA_FLG and D_UNMTCH. Thus , in this query condition " NVL (I.MA_FLG, 'N') <> 'Y' results in ~105million and NVL (I.D_UNMTCH, 'N') <> 'Y' results in ~111million. So how should I create index or modify code to make it the best access/filter criteria so as to make the query faster? MA_FLG Count(*) N 105228656 Y 6000938 643566 D_UNMTCH Count(*) Y 13715 111859445 On Wed, Aug 25, 2021 at 8:00 PM Mark W. Farnham <mwf_at_rsiz.com> wrote: The other thing, for flag values like AND NVL (I.MA_FLG, 'N') <> 'Y' AND NVL (I.D_UNMTCH, 'N') <> 'Y' if you’re thinking about adding an index, and even if you need a virtual column to do this because you have too much code depending on values ‘N’ and ‘Y’, define the final status (the one where nearly all of them land) as NULL, being the ones you are NOT interested in most of the time. In both these cases it looks like ‘Y’ would then be NULL, so i.ma_flg_v is defined decoding Y to NULL and anything else to N and your code becomes and i.ma_flg_v = ‘N’ and you deal with variability in non-nulls that are not ‘Y’ on the original, or i.ma_flg_v decodes Y to NULL, NULL to ‘N’ and anything else unchanged and your code becomes i.ma_flg_v is NOT NULL, or you make a functional index on i.ma_flg that does the equivalent. I can’t remember off the top of my head whether either way gives you a real advantage over the other in stats collections and the CBO doing something smart and that probably changed over the releases. That might be in one of my papers. When you then index that column the nulls disappear, leaving you with a very tiny index to prune your result set immediately to very small and you can usually filter the rest fast without an index. Remember, ORACLE cannot assign a value to NULL in anything they do. But YOU can. When this is appropriate, it is one of the neatest and easiest “magic tricks” in the Oracle kit. Good luck, mwf
From:oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Sayan Malakshinov
Sent: Wednesday, August 25, 2021 9:40 AM
To: Lok P
Cc: Oracle L
Subject: Re: Fixing Performance issue with less selective columns
Hi Lok, > SUBSTR(:B8,0.50) Looks like this query should be analyzed and tested better. You haven't provided histograms and bind values statistics, so not enough info to analyze it properly. For now it looks like "I.WOF_DATE IS NULL" is one of the most selective predicates - it gives only 83154 nulls. In addition to histogram statistics(dba_tab_histograms) and most often binds values, I would like also to see what does return this query:
select
NVL(I.MA_FLG, 'N'),NVL(I.D_UNMTCH, 'N'),I.DC_CODE,count(*)
FROM PP_IN_TAB I
group by NVL(I.MA_FLG, 'N'),NVL(I.D_UNMTCH, 'N'),I.DC_CODE;
On Wed, Aug 25, 2021 at 4:14 PM Lok P <loknath.73_at_gmail.com> wrote:
Hello , This database has version 11.2.0.4 of Oracle. We have the below query which is executed thousands of times. It's used in a plsql function which in turn gets called from a procedure. And this procedure gets called from java thousands of times. And I see from dba_hist_sqlstat , for most of the runs this below query results in zero rows. We see from the active session history for the overall process this query is consuming most time/resources and making the process run longer. So wanted to understand if we can make this individual query execution faster which would eventually make the process faster? The base table- PP_IN_TAB is holding ~111million rows and is ~43GB in size. Column PP_ID is the primary key here. The filter predicates used in this query are below. Many of them were not very selective in nature. So I am not able to conclude if any composite index is going to help us here. Can you please guide me , what is the correct approach to tune this process in such a scenario? Below is the column data pattern used as filter predicate in this query. Most of these are less selective in nature.
TABLE_NAME COLUMN_NAME NUM_DISTINCT NUM_NULLS PP_IN_TAB EF_ID 39515 6151686 PP_IN_TAB PE 103074806 647050 PP_IN_TAB PT_Code 24 0 PP_IN_TAB PT_MCODE 20 0 PP_IN_TAB D_CUR_CODE 13 592784 PP_IN_TAB ED_AMT 320892 6 PP_IN_TAB WOF_DATE 2572 83154 PP_IN_TAB PR_CTGRY 2 86 PP_IN_TAB PDE_RSN_CAT 6 0 PP_IN_TAB MA_FLG 2 648172 PP_IN_TAB M_TXT 29460248 9118572 PP_IN_TAB D_UNMTCH 1 111766716
SELECT NVL (I.PP_ID, 0)
FROM PP_IN_TAB I
WHERE TRIM(I.M_TXT) = TRIM (SUBSTR ( :B8, 0.50)) AND I.PT_Code = :B7 AND NVL ( :B6, I.PT_MCODE) = NVL ( :B6, :B5) AND I.DC_CODE = :B4 AND I.D_CUR_CODE = :B3 AND I.ED_AMT = :B2 AND I.PR_CTGRY = :B1 AND I.PE IS NOT NULL AND I.EF_ID IS NULL AND I.WOF_DATE IS NULL AND NVL (I.MA_FLG, 'N') <> 'Y' AND NVL (I.D_UNMTCH, 'N') <> 'Y' AND ROWNUM = 1;
Global Information
Status : DONE (ALL ROWS) Instance ID : 1 SQL Execution ID : 16777216 Execution Started : 08/25/2021 03:53:25 First Refresh Time : 08/25/2021 03:53:25 Last Refresh Time : 08/25/2021 03:53:28 Duration : 3s Module/Action : SQL*Plus/- Program : sqlplus.exe Fetch Calls : 1Global Stats
| Elapsed | Cpu | IO | Application | Fetch | Buffer | Read | Read | Cell | | Time(s) | Time(s) | Waits(s) | Waits(s) | Calls | Gets | Reqs | Bytes | Offload |
| 3.30 | 1.15 | 2.15 | 0.00 | 1 | 6M | 44379 | 43GB | 99.99% |
SQL Plan Monitoring Details (Plan Hash Value=1096440065)
| Id | Operation | Name | Rows | Cost | Time | Start | Execs | Rows | Read | Read | Cell | Mem | Activity | Activity Detail | | | | | (Estim) | | Active(s) | Active | | (Actual) | Reqs | Bytes | Offload | (Max) | (%) | (# samples) |Predicate Information (identified by operation id):
==========================================================================================================================================================================================
| 0 | SELECT STATEMENT | | | | | | 1 | | | | | | | | | 1 | COUNT STOPKEY | | | | | | 1 | | | | | | | | | 2 | TABLE ACCESS STORAGE FULL | PP_IN_TAB | 1 | 128K | 3 | +2 | 1 | 0 | 44379 | 43GB | 99.99% | 6M | 100.00 | cell smart table scan (3) |
==========================================================================================================================================================================================
1 - filter(ROWNUM=1)
2 - storage("I"."WOF_DATE" IS NULL AND "I"."EF_ID" IS NULL AND "I"."PT_Code"=:B7 AND "I"."D_CUR_CODE"=:B3 AND "I"."PR_CTGRY"=:B1 AND
"I"."DC_CODE"=:B4 AND "I"."ED_AMT"=TO_NUMBER(:B2) AND NVL(:B6,"I"."PT_MCODE")=NVL(:B6,:B5) AND TRIM("I"."M_TXT")=TRIM(SUBSTR(:B8,0.50)) AND "I"."PE" IS NOT NULL AND NVL("I"."MA_FLG",'N')<>'Y' AND NVL("I"."D_UNMTCH",'N')<>'Y') filter("I"."WOF_DATE" IS NULL AND "I"."EF_ID" IS NULL AND "I"."PT_Code"=:B7 AND "I"."D_CUR_CODE"=:B3 AND "I"."PR_CTGRY"=:B1 AND "I"."DC_CODE"=:B4 AND "I"."ED_AMT"=TO_NUMBER(:B2) AND NVL(:B6,"I"."PT_MCODE")=NVL(:B6,:B5) AND TRIM("I"."M_TXT")=TRIM(SUBSTR(:B8,0.50)) AND "I"."PE" IS NOT NULL AND NVL("I"."MA_FLG",'N')<>'Y' AND NVL("I"."D_UNMTCH",'N')<>'Y') --
Best regards,
Sayan Malakshinov
Oracle performance tuning engineer
Oracle ACE Associate
http://orasql.org
-- http://www.freelists.org/webpage/oracle-lReceived on Thu Aug 26 2021 - 10:00:28 CEST