Tales From A Lazy Fat DBA

Love all databases! – Its all about performance, troubleshooting & much more …. ¯\_(ツ)_/¯

  • Prashant Dixit is the 'FatDBA' ...
  • Follow me on Twitter

A ‘simple’ dissection of an optimizer decision and cost using Oracle 10053 trace event … Yes, it’s simple!

Posted by FatDBA on October 10, 2021

Hi Guys,

In today’s post I will try to explain the cost, stats (system & object level), join order, cardinality & the selectivity calculations about execution plan entries and will take help of the infamous 10053 optimizer traces. Though 10053 is notoriously complicated, detailed and extensive even for simple SQL statements but always has the wealth of information about optimizer decisions like plan directives, peeked bind values, predicate move around, Join elimination stats, query transformations, fix controls, join eliminations, join orders (easily complicated) & other computations or calculations. So, I will try to keep it as simple as I can.

About ways of generating the report, you can access it via ALTER SESSION, ORADEBUG or even through DBMS_SQLDIAG.DUMP_TRACE (does not require diagnostic pack). There are multiple use cases of using 10053, but the most common is when you want to compare any best or worst plans and you want to see the difference in cost calculations.

Alright, so this is the complete SQL statement followed up by its execution plan and predicate details.

SELECT * FROM (SELECT n.object_inst_id, n.name, n.type, rownum idx
FROM testsch.PDI_FBI_HASHAH1 n join testsch.PDI_FBI_HASHAH1_b on  TESTID_name = :"SYS_B_00"  and owning_object_id = n.object_inst_id
and n.name = NVL(null, n.name) AND n.status = NVL(:"SYS_B_01", :"SYS_B_02")
AND n.type = NVL(:"SYS_B_03", n.type) AND NVL(n.test_allocation,:"SYS_B_04") = NVL(NVL(null, n.test_allocation),:"SYS_B_05")
AND  NVL(n.number_usage, :"SYS_B_06")=:"SYS_B_07" AND ( n.owner IS NULL OR instr(lower(n.owner), :"SYS_B_08") > :"SYS_B_09" )
AND n.assigned_csn_id IS NULL AND n.locked_on IS NULL AND n.lock_owner IS NULL
AND (n.lock_id IS NULL OR n.LOCK_DATE <= (sysdate - NVL(n.lock_period,
(SELECT dixi_trait_values FROM testsch.PDI_HASHIS_MERGEDINFO_DATA WHERE HASH_BATTR_SHA = :"SYS_B_10")) / :"SYS_B_11" / :"SYS_B_12"))
AND target_object_id = :"SYS_B_13" ORDER BY TO_NUMBER(n.name)) OUTPUTS WHERE OUTPUTS.idx between :"SYS_B_14" and :"SYS_B_15"


----------------------------------------------------------------------------------------------------------------
| Id  | Operation                      | Name                          | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT               |                               |     2 |   364 |   107   (0)| 00:00:01 |
|*  1 |  VIEW                          |                               |     2 |   364 |   107   (0)| 00:00:01 |
|   2 |   COUNT                        |                               |       |       |            |          |
|   3 |    NESTED LOOPS                |                               |     2 |   340 |   107   (0)| 00:00:01 |
|*  4 |     TABLE ACCESS BY INDEX ROWID| PDI_FBI_HASHAH1               |  6719 |   505K|   106   (0)| 00:00:01 |
|*  5 |      INDEX FULL SCAN           | PDI_FBI_NNAME_UTYPE_STA_NPFIX |  3065 |       |     7   (0)| 00:00:01 |
|*  6 |      TABLE ACCESS FULL         | PDI_HASHIS_MERGEDINFO_DATA    |     1 |    28 |     3   (0)| 00:00:01 | --> How the CBO decided cost 3 ? 
|*  7 |     INDEX RANGE SCAN           | IDX_TRNNUMBERB_SMOKE          |     1 |    93 |     1   (0)| 00:00:01 |
----------------------------------------------------------------------------------------------------------------


Predicate Information (identified by operation id):
---------------------------------------------------
   1 - filter("OUTPUTS"."IDX"<=10 AND "OUTPUTS"."IDX">=1)
   4 - filter("N"."TYPE"='MOBILE' AND ("N"."OWNER" IS NULL OR INSTR(LOWER("N"."OWNER"),'dna oyj')>0)
              AND NVL("N"."NUMBER_USAGE",'External')='Internal' AND "N"."ASSIGNED_CSN_ID" IS NULL AND
              NVL("N"."TEST_ALLOCATION",'EAI')=NVL(NVL(NULL,"N"."TEST_ALLOCATION"),'EAI') AND "N"."LOCKED_ON" IS NULL
              AND "N"."LOCK_OWNER" IS NULL AND ("N"."LOCK_ID" IS NULL OR
              "N"."LOCK_DATE"<=SYSDATE@!-NVL("N"."LOCK_PERIOD", (SELECT "DIXI_TRAIT_VALUES" FROM
              "TESTSCH"."PDI_HASHIS_MERGEDINFO_DATA" "PDI_HASHIS_MERGEDINFO_DATA" WHERE
              "HASH_BATTR_SHA"='MY_CALL_DURATION'))/60/24))
   5 - access("N"."STATUS"='Available')
       filter("N"."STATUS"='Available')
   6 - filter("HASH_BATTR_SHA"='MY_CALL_DURATION')
   7 - access("TESTID_NAME"='ind.fatdba.application.dss.more.production.BESTID.PDINumberBestSubClass'
               AND "OWNING_OBJECT_ID"="N"."OBJECT_INST_ID" AND "TARGET_OBJECT_ID"=21)

Let’s try to understand how optimizer come up with the cost numbers, how it derives these values, on what basis and what all it considers during the calculations. I will explaining about one of the access operation at Line ID 6 of the plan where it’s doing a TABLE ACCESS FULL on table PDI_HASHIS_MERGEDINFO_DATA and has a final cost of 3 associated with it.

So, what it is ? what all was happened under the hood by the optimizer to get this value of 3 for this operation ?

This is the subquery for which we will be checking the cost calculations within 10053 trace files.

(SELECT dixi_trait_values FROM testsch.PDI_HASHIS_MERGEDINFO_DATA WHERE HASH_BATTR_SHA = :"SYS_B_10")) / :"SYS_B_11" / :"SYS_B_12"))

Let me replace the sys generated values with their original values, for this I will be checking under “Peeked values of the binds in SQL statement” section. The same you can get from ‘predicate information’ section of the execution plan too, but let’s stick to 10053 for the moment.

*******************************************
Peeked values of the binds in SQL statement
*******************************************
----- Bind Info (kkscoacd) -----
 Bind#0
  oacdty=01 mxl=128(73) mxlc=00 mal=00 scl=00 pre=00
  oacflg=10 fl2=0100 frm=01 csi=873 siz=128 off=0
  kxsbbbfp=7f3791745ee0  bln=128  avl=73  flg=09
  value="ind.fatdba.application.dss.more.production.BESTID.PDINumberBestSubClass"
...
......
........

 Bind#10
  oacdty=01 mxl=32(19) mxlc=00 mal=00 scl=00 pre=00
  oacflg=10 fl2=0100 frm=01 csi=873 siz=32 off=0
  kxsbbbfp=7f3791745d48  bln=32  avl=19  flg=09
  value="MY_CALL_DURATION"
 Bind#11
  oacdty=02 mxl=22(02) mxlc=00 mal=00 scl=00 pre=00
  oacflg=10 fl2=0100 frm=00 csi=00 siz=24 off=0
  kxsbbbfp=7f3791745d20  bln=22  avl=02  flg=09
  value=60
 Bind#12
  oacdty=02 mxl=22(02) mxlc=00 mal=00 scl=00 pre=00
  oacflg=10 fl2=0100 frm=00 csi=00 siz=24 off=0
  kxsbbbfp=7f3791745cf8  bln=22  avl=02  flg=09
  value=24

Okay, let’s place them in the subquery.

SELECT dixi_trait_values FROM testsch.PDI_HASHIS_MERGEDINFO_DATA WHERE HASH_BATTR_SHA='MY_CALL_DURATION/60/24';

These are the contents of the column.

SQL> SELECT dixi_trait_values FROM testsch.PDI_HASHIS_MERGEDINFO_DATA;

DIXI_TRAIT_VALUES
-----------------
             60
            120
            200
             10
             60

Let’s check for access path analysis for this SQL.

***************************************
BASE STATISTICAL INFORMATION
***********************
Table Stats::
  Table: PDI_HASHIS_MERGEDINFO_DATA  Alias: PDI_HASHIS_MERGEDINFO_DATA
  #Rows: 5  SSZ: 0  LGR: 0  #Blks:  5  AvgRowLen:  31.00  NEB: 0  ChainCnt:  0.00  SPC: 0  RFL: 0  RNF: 0  CBK: 0  CHR: 0  KQDFLG: 1
  #IMCUs: 0  IMCRowCnt: 0  IMCJournalRowCnt: 0  #IMCBlocks: 0  IMCQuotient: 0.000000
try to generate single-table filter predicates from ORs for query block SEL$3 (#0)
finally: "PDI_HASHIS_MERGEDINFO_DATA"."DIXI_TRAIT_VALUES"=:B1

=======================================
SPD: BEGIN context at query block level
=======================================
Query Block SEL$3 (#0)
Applicable DS directives:
Return code in qosdSetupDirCtx4QB: NODIR
=====================================
SPD: END context at query block level
=====================================
Access path analysis for PDI_HASHIS_MERGEDINFO_DATA
***************************************
SINGLE TABLE ACCESS PATH 
  Single Table Cardinality Estimation for PDI_HASHIS_MERGEDINFO_DATA[PDI_HASHIS_MERGEDINFO_DATA] 
  SPD: Return code in qosdDSDirSetup: NOQBCTX, estType = TABLE
  Column (#1): CONF_ATTR_NAME(VARCHAR2)
    AvgLen: 24 NDV: 5 Nulls: 0 Density: 0.200000
  Table: PDI_HASHIS_MERGEDINFO_DATA  Alias: PDI_HASHIS_MERGEDINFO_DATA
    Card: Original: 5.000000  Rounded: 1  Computed: 1.000000  Non Adjusted: 1.000000
  Scan IO  Cost (Disk) =   3.000000
  Scan CPU Cost (Disk) =   36377.200000
  Total Scan IO  Cost  =   3.000000 (scan (Disk))
                         + 0.000000 (io filter eval) (= 0.000000 (per row) * 5.000000 (#rows))
                       =   3.000000
  Total Scan CPU  Cost =   36377.200000 (scan (Disk))
                         + 250.000000 (cpu filter eval) (= 50.000000 (per row) * 5.000000 (#rows))
                       =   36627.200000
  Access Path: TableScan
    Cost:  3.002419  Resp: 3.002419  Degree: 0
      Cost_io: 3.000000  Cost_cpu: 36627
      Resp_io: 3.000000  Resp_cpu: 36627
  Best:: AccessPath: TableScan
         Cost: 3.002419  Degree: 1  Resp: 3.002419  Card: 1.000000  Bytes: 0.000000

.....
.......

OPTIMIZER STATISTICS AND COMPUTATIONS
PJE:  Bypassed; QB has a single table SEL$3 (#0)
***************************************
GENERAL PLANS
***************************************
Considering cardinality-based initial join order.
Permutations for Starting Table :0
Join order[1]:  PDI_HASHIS_MERGEDINFO_DATA[PDI_HASHIS_MERGEDINFO_DATA]#0
***********************
Best so far:  Table#: 0  cost: 3.002419  card: 1.000000  bytes: 28.000000
***********************

If you check above you will see it considers the CPU & IO to generate the final cost values and that’s what we already know. Cost is the amount of work the optimizer estimates to run your query via a specific plan. The optimizer generally prefers lower cost plans. The optimizer’s cost model accounts for the IO, CPU, and network resources that will be used by the query.
IO_COST: I/O cost of the operation as estimated by the query optimizer’s approach. The value of this column is proportional to the number of data blocks read by the operation. For statements that use the rule-based approach, this column is null.
CPU_COST: CPU cost of the operation as estimated by the query optimizer’s approach. The value of this column is proportional to the number of machine cycles required for the operation. For statements that use the rule-based approach, this column is null.

At the same time it checks for dictionary stats to get table level information i.e. average row lenght, blocks, total rows and takes help of metrics NDV (nuber of distinct values), nulls, density of rows etc. to calculate the original cardinality of 5 (number of distinct entries) and the rounded cardinality of 1 – this SQL has resulted only into 1 row as an output, hence 1 is the rounded cardinality.

So, the main elements are ‘Scan IO Cost (Disk)’, ‘Scan CPU Cost (Disk)’ which finally derives the ‘Total Scan IO Cost’ value of 3.

That’s how the optimizer has calculated this value of 3 for this specific access operation, similarly you understand any of the other access entries in plan, and their associated costs.

If you want to read more on 10053 traces, I recommend you a great article by Wolfgang Breitling, Jonathan Lewis (his book cost-based Oracle fundamentals) & few posts from Maria Colgan.

Hope It Helped
Prashant Dixit

One Response to “A ‘simple’ dissection of an optimizer decision and cost using Oracle 10053 trace event … Yes, it’s simple!”

  1. […] am sure that my last post about 10053 debug traces has sparked some interest in optimizer cost calculations and estimations […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

 
%d bloggers like this: