Melbourne Groundbreakers Tour - Upgrading without risk

Copyright © 2017, Oracle and/or its affiliates. All rights reserved.
Taming the beast
The 12c Optimizer
Connor McDonald
1

Connor McDonald

3

4

Copyright © 2018, Oracle and/or its affiliates. All rights reserved. |
Stuff
youtube bit.ly/youtube-connor
blog bit.ly/blog-connor
twitter bit.ly/twitter-connor
400+ posts mainly on database & development
250 technical videos, new uploads every week
rants and raves on tech and the world :-)

etc...
facebook bit.ly/facebook-connor
linkedin bit.ly/linkedin-connor
instagram bit.ly/instagram-connor
slideshare bit.ly/slideshare-connor

7https://asktom.oracle.com

https://asktom.oracle.com/officehours

150 hours free access so far
9

this session
10

3 things
11

12c optimizer …
better performance
12

12c optimizer …
worse performance
13

worse performance …
14

15

16

goal of the optimizer
17

simple
18

good execution plan
19

so how did we end up here ?
20

21
You will upgrade us to 12c
with its new optimizer

LOL !LOL !
22

LMAOLMAO!
23

That dude
is toast!
That dude
is toast!
24

Is this why I don't
have a desk ?
25

lots of fear
26

lots of mis-information
27

only 2 issues
28

ISSUE #1
29

missing or poor statistics
30

cardinality mis-estimates
31

cardinality is everything
32

33

better stats
34

35

1) better histograms
36

2015
37

38

39

Histograms are so
awesome. See
how much they
make me smile !
40

41

No …
… honest :-)
42

the sucky part
43

SQL> create table T
2 as select * from dba_objects;
Table created.
SQL> exec dbms_stats.gather_table_stats('','T');
PL/SQL procedure successfully completed.
44

SQL> select COLUMN_NAME,NUM_DISTINCT,
2 ( select count(*)
3 from all_tab_histograms
4 where owner = a.owner
5 and table_name = a.table_name
6 and column_name = a.column_name )
7 from all_tab_cols a
8 where table_name = 'T'
9 order by column_id;
COLUMN_NAME NUM_DISTINCT HIST_CNT
------------------------------ ------------ ----------
OWNER 25 2
OBJECT_NAME 53000 2
SUBOBJECT_NAME 230 2
OBJECT_ID 90808 2 45

46

47

SQL> select intcol#,equality_preds,
2 range_preds,timestamp
3 from sys.col_usage$
4 where obj# = (select object_id
5 from obj
6 where object_name = 'T' );
INTCOL# EQUALITY_PREDS RANGE_PREDS TIMESTAMP
---------- -------------- ----------- ---------
1 1 0 24-SEP-17
48

SQL> select COLUMN_NAME,NUM_DISTINCT,
2 ( select count(*)
3 from all_tab_histograms
4 where owner = a.owner
5 and table_name = a.table_name
6 and column_name = a.column_name )
7 from all_tab_cols a
8 where table_name = 'T'
9 order by column_id;
COLUMN_NAME NUM_DISTINCT HIST_CNT
------------------------------ ------------ ----------
OWNER 25 25
OBJECT_NAME 53000 2
SUBOBJECT_NAME 230 2
OBJECT_ID 90808 2
49

we don't know what your "app" is
50

still requires care
51

12c improvements
52

SQL> select status, count(*)
2 from orders
3 group by status
4 order by 1;
STATUS COUNT(*)
-------------------- ----------
1-New 1000
2-Assigned 10
3-InProgress 4000
4-Processed 3
5-Shipped 800
6-Received 5000
7-Completed 15000
8-Archived 20000
53

skewed ...
54

... need histogram
55

11g
56

SQL> begin
2 dbms_stats.gather_table_stats('','ORDERS',
3 method_opt=>'for all columns size 15');
4 end;
5 /
SQL> select histogram from USER_TAB_COL_STATISTICS
2 where table_name = 'ORDERS';
HISTOGRAM
---------------
FREQUENCY
57

SQL> select
2 endpoint_number,
3 substr(
4 utl_raw.cast_to_varchar2(
5 hextoraw(to_char(endpoint_value,rpad('FM',65,'X')))),1,5),
6 endpoint_number -
7 nvl(lag(endpoint_number) over ( order by endpoint_number),0)
8 from user_histograms
9 where table_name = 'ORDERS'
10 order by 1;
ENDPOINT_NUMBER END_VAL FREQ
--------------- -------------------- ----------
1000 1-Nev 1000
1010 2-Ass 10
5010 3-InP 4000
5013 4-Pro 3
5813 5-Shi 800
10813 6-Rec 5000
25813 7-Com 15000
45813 8-Arc 20000
58

SQL> select
2 endpoint_number,
3 substr(
6 endpoint_number -
10 order by 1;
--------------- -------------------- ----------
1000 1-Nev 1000
1010 2-Ass 10
5010 3-InP 4000
5013 4-Pro 3
5813 5-Shi 800
10813 6-Rec 5000
25813 7-Com 15000
45813 8-Arc 20000
59

SQL> select
2 endpoint_number,
3 substr(
6 endpoint_number -
10 order by 1;
--------------- -------------------- ----------
1000 1-Nev 1000
1010 2-Ass 10
5010 3-InP 4000
5013 4-Pro 3
5813 5-Shi 800
10813 6-Rec 5000
25813 7-Com 15000
45813 8-Arc 20000
60

SQL> select * from orders
2 where status = '5-Shipped';
----------------------------------------------------
| Id | Operation | Name | Rows | Bytes |
----------------------------------------------------
| 0 | SELECT STATEMENT | | 800 | 8800 |
|* 1 | TABLE ACCESS FULL| ORDERS | 800 | 8800 |
----------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter("STATUS"='5-Shipped')
61
true = 800

so far … so good
62

NDV > 254*
63

SQL> begin
4 end;
5 /
HISTOGRAM
---------------
HEIGHT BALANCED
64

SQL> select
2 endpoint_number,
3 substr(
6 endpoint_number -
10 order by 1;
ENDPOINT_NUMBER END_VAL BKTS
--------------- -------------------- ----------
0 1-Nev 0
1 6-Rec 1
3 7-Com 2
6 8-Arc 3
65

1-New
6-Received
7-Complete
8-Archived
8-Archived
8-Archived
7-Complete
66

2 where status = '8-Archived';
----------------------------------------------------
----------------------------------------------------
| 0 | SELECT STATEMENT | | 19089 | 205K|
|* 1 | TABLE ACCESS FULL| ORDERS | 19089 | 205K|
----------------------------------------------------
---------------------------------------------------
1 - filter("STATUS"='8-Archived')
true = 20000
67

2 where status = '7-Completed';
-----------------------------------------------------
-----------------------------------------------------
| 0 | SELECT STATEMENT | | 15271 | 164K|
|* 1 | TABLE ACCESS FULL| ORDERS | 15271 | 164K|
-----------------------------------------------------
---------------------------------------------------
1 - filter("STATUS"='7-Completed')
true = 10000
68

69

1-New
6-Received
7-Complete
8-Archived
8-Archived
8-Archived
7-Completetotal ~ 45,000 ~7000 per bucket 70

1-New
6-Received
7-Complete
8-Archived
8-Archived
8-Archived
7-Completetotal = 45,000 ~7000 per bucket 71

2 where status = '2-Assigned';
----------------------------------------------------
----------------------------------------------------
----------------------------------------------------
---------------------------------------------------
1 - filter("STATUS"='2-Assigned')
true = 3
72

height balanced means ...
73

bucket spanning = dramas
74

bucket swallowing = dramas
75

11g
HEIGHT-BALANCED
FREQUENCY
76

Footnote: 11g
FOR ALL COLUMNS SIZE 25AUTO
77

78

SQL> begin
4 end;
5 /
HISTOGRAM
---------------
FREQUENCY
79

SQL> begin
4 end;
5 /
HISTOGRAM
---------------
TOP-FREQUENCY
80

SQL> select
2 endpoint_number,
3 substr(
5 hextoraw(to_char(endpoint_value,rpad('FM',65,'X')))),1,5)
6 endpoint_number -
10 order by 1;
--------------- -------------------- ----------
1000 1-Nev 1000
5000 3-InP 4000
5800 5-Shi 800
10800 6-Rec 5000
25800 7-Com 15000
45800 8-Arc 20000 81

1-New
6-Received
3-InProgress
5-Shipped
8-Archived
7-Completed
82

other values ?
83

SQL> select num_rows
2 from user_tables
NUM_ROWS
----------
45813
SQL> select num_distinct
2 from user_tab_cols
NUM_DISTINCT
------------
8
FREQ
----------
800
1000
4000
5000
15000
20000
========
45800
( 45813 - 45800 ) / ( 8 - 6 ) =
84

2 where status = '4-Processed';
-----------------------------------------------------
-----------------------------------------------------
-----------------------------------------------------
---------------------------------------------------
1 - filter("STATUS"='4-Processed')
85
true = 3

what defines "TOP"
% top values > ((bkt-1)/bkt)* num
eg 1000 rows/ 12 bucket
12 most frequent > 11/12 * 1000
~920 rows
86

what if TOP FREQ not possible ?
87

SQL> insert into orders select '1-New'
2 from dual connect by level <= 9000;
SQL> insert into orders select '2-Assigned'
SQL> insert into orders select '3-InProgress'
SQL> insert into orders select '4-Processed'
SQL> insert into orders select '5-Shipped'
SQL> insert into orders select '6-Received'
88

SQL> select status, count(*)
2 from orders
3 group by status
4 order by 1;
STATUS COUNT(*)
-------------------- ----------
1-New 10000
2-Assigned 9510
3-InProgress 10000
4-Processed 9503
5-Shipped 9600
6-Received 9500
7-Completed 15000
8-Archived 20000
89

SQL> begin
4 end;
5 /
HISTOGRAM
---------------
HYBRID
90

SQL> select
2 endpoint_number,
3 substr(
6 endpoint_number -
7 nvl(lag(endpoint_number) over ( order by endpoint_number),0),
8 endpoint_repeat_count
11 order by 1;
ENDPOINT_NUMBER END_VAL FREQ ENDPOINT_REPEAT_COUNT
--------------- -------------------- ---------- ---------------------
590 1-Nev 590 590
1743 3-InP 1153 626
2314 4-Pro 571 571
2895 5-Shi 581 581
3449 6-Rec 554 554
5547 8-Arc 2098 1213 91

1-New
6-Received
3-InProgress
5-Shipped
8-Archived
4-Processed
ENDPOINT_NUMBER END_VAL FREQ ENDPOINT_REPEAT_COUNT
--------------- -------------------- ---------- ---------------------
590 1-Nev 590 590
1743 3-InP 1153 626
2314 4-Pro 571 571
2895 5-Shi 581 581
3449 6-Rec 554 554
5547 8-Arc 2098 1213
2-Assigned
590 1153 571 581 554 2098
7-Completed
92

8-Archived
END_VAL FREQ ENDPOINT_REPEAT_COUNT
---------------- -------- ---------------------
1-Nev 590 590
3-InP 1153 626
4-Pro 571 571
5-Shi 581 581
6-Rec 554 554
8-Arc 2098 1213
2098
7-Completed
"Bucket 8-Arc has 2098 occurrences...
1213 of them are the end value 8-Arc...
therefore 885 are > 6-Rec and < 8-Arc"
93
6-Received
554

no bucket spanning
94

key point
95

must use auto sample size
96

still need care
97

representative values
98

"today's hot deals"
99

gather statistics
SQL> select count(*)
2 from items
3 where type = 'HOT_DEAL'
COUNT(*)
----------
0
100

activate "deals of the day"
SQL> update items
2 set type = 'HOT_DEAL'
3 where ...
SQL> select count(*)
2 from items
COUNT(*)
----------
14
101

biggest day of the year !
SQL> select min(discount)
2 from items
4 and ...
where type = 'HOT_DEAL'
SQL> select free_shipping
2 from items
3 where hot_deal = 'Y'
4 and ...
SQL> select ...
2 from ...
SQL> select ...
2 from ...
SQL> select ...
2 from ...
SQL> select ...
2 from ...SQL> select ...
2 from ...
102

where type = 'HOT_DEAL'
103

104

SQL> select type, count(*)
2 from items
3 group by type;
TYPE COUNT(*)
------ ----------
ORDERED 21353
ARCHIVED 4213
NORMAL 513234
SOLD 528291
HOT_DEAL 0
105

normally...
106

"ok... I got 1,000,000 total rows...
and 7 distinct values.
I got a frequency histogram...
There's 4 popular values,
adding up to 910,000
That's 3 values left to cover the rest
3 / ( 1,000,000 - 910,000) = density
... I'm done!"
107

SINGLE TABLE ACCESS PATH
Single Table Cardinality Estimation for ITEMS[ITEMS]
SPD: Return code in qosdDSDirSetup: NOCTX, estType = TABLE
kkecdn: Single Table Predicate:"ITEMS"."TYPE"='HOT_DEAL'
Column (#2):
NewDensity:0.001974, OldDensity:0.000000 BktCnt:1067091.000000,
PopBktCnt:1067091.000000, PopValCnt:4, NDV:4
Column (#2): TYPE(VARCHAR2)
AvgLen: 7 NDV: 4 Nulls: 0 Density: 0.001974
Histogram: Freq #Bkts: 4 UncompBkts: 1067091 EndPtVals: 4
Using density:
0.001974 of col #2 as selectivity of pred having unreasonably low value
NDV / (1067091 - 1067091)"Aaggghhhh!"
108

SQL> select * from items where type = 'HOT_DEAL';
----------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)|
----------------------------------------------------------------
| 0 | SELECT STATEMENT | | 2107 | 23177 | 672 (2)|
|* 1 | TABLE ACCESS FULL| ITEMS | 2107 | 23177 | 672 (2)|
----------------------------------------------------------------
TYPE COUNT(*)
------ ----------
ORDERED 21353
ARCHIVED 4213 * 50%
NORMAL 513234
SOLD 528291
109

representative values
... at the right time
110

2) extensions
111

11g
112

"extended" statistics
113

aka "column groups"
114

solve GIGO
115

SQL> select *
2 from ADDRESS
3 where CITY = 'Bangalore'
4 and COUNTRY = 'Australia';
116

117

automatic column groups
118

12.2
119

expression tracking
120

SQL> select s.item_id,
2 s.category,
3 p.prod_list_price - p.prod_min_price
4 from products p,
5 sales s
6 where ....;
SQL> select expression_text, evaluation_count, fixed_cost
2 from user_expression_statistics
2 where table_name = 'PRODUCTS';
EXPRESSION_TEXT EVALUATION_COUNT FIXED_COST
---------------------------------- ---------------- ----------
"PROD_LIST_PRICE"-"PROD_MIN_PRICE" 766 .000041667
121

122

3) online gather
123

the old problem
124

SQL> insert /*+ APPEND */ into MY_TABLE
2 select *
3 from MY_WHOPPING_GREAT_FAT_TABLE;MY_WHOPPING_GREAT_FAT_TABLE;
125

126

127

128

129

130

SQL> insert /*+ APPEND */ into MY_TABLE
2 select *
3 from MY_WHOPPING_GREAT_FAT_TABLE
4 ...
7102984123 rows created.
Elapsed: 06:12:34.00
131

and then...
132

SQL> select num_rows from user_tables
2 where table_name = 'MY_TABLE';
NUM_ROWS
----------
133

so ...
134

SQL> begin
2 dbms_stats.gather_table_stats('',
3 tname=>'MY_TABLE'
4 ...
5 ...
6 end;
7 /
Elapsed: a bloody long time
135

136

on load
137

SQL> create table T (
2 ...
3 ...
4 ... );
Table created.
2 from user_tables
3 where table_name = 'T';
NUM_ROWS
----------
SQL> insert /*+ APPEND */ into T
2 select * from dba_objects;
78876 rows created.
SQL> commit;
Commit complete.
2 from user_tables
NUM_ROWS
----------
78876 138

things to note
139

must be direct
insert /*+ APPEND */
create table as select
140

SQL> insert into T select * from dba_objects;
78876 rows created.
SQL> commit;
NUM_ROWS
----------
141

must be empty
142

SQL> insert into T values (....)
1 row created.
SQL> insert /*+ APPEND */ into T select * from dba_objects;
78876 rows created.
SQL> commit;
NUM_ROWS
----------
143

execution plan
144

SQL> insert /*+ APPEND */ into T
2 select * from t_src;
------------------------------------------------------------------------------
------------------------------------------------------------------------------
| 0 | INSERT STATEMENT | | 78877 | 9M| 428 (1)|
| 1 | LOAD AS SELECT | T | | | |
| 2 | OPTIMIZER STATISTICS GATHERING | | 78877 | 9M| 428 (1)|
| 3 | TABLE ACCESS FULL | T_SRC| 78877 | 9M| 428 (1)|
------------------------------------------------------------------------------
145

SQL> select column_name, notes
2 from user_tab_col_statistics
COLUMN_NAME NOTES
------------------------------ -------------
OWNER STATS_ON_LOAD
OBJECT_NAME STATS_ON_LOAD
SUBOBJECT_NAME STATS_ON_LOAD
OBJECT_ID STATS_ON_LOAD
DATA_OBJECT_ID STATS_ON_LOAD
OBJECT_TYPE STATS_ON_LOAD
CREATED STATS_ON_LOAD
...
...
146

note: no histograms
147

3) session GTT
148

11g
149

SQL> create global temporary table t ( x int, y int )
2 on commit preserve rows;
Table created.
SQL> insert into t
2 select rownum, rownum
3 from dual
4 connect by level <= 10000;
10000 rows created.
150

SQL> set autotrace traceonly explain
SQL> select * from t;
---------------------------------------------------------------
---------------------------------------------------------------
| 1 | TABLE ACCESS FULL| T | 10000 | 80000 | 7 (0)|
---------------------------------------------------------------
151

looks great until ...
152

SQL> connect scott/tiger
Connected.
SQL> select count(*) from t;
COUNT(*)
----------
0
---------------------------------------------------------------
---------------------------------------------------------------
---------------------------------------------------------------
153

154

SQL> create global temporary table t ( x int, y int )
2 on commit preserve rows;
Table created.
SQL> insert into t
3 from dual
10000 rows created.
155

---------------------------------------------------------------
---------------------------------------------------------------
---------------------------------------------------------------
as before
156

SQL> connect scott/tiger
Connected.
SQL> insert into t
3 from dual
50 rows created.
157

---------------------------------------------------------------
---------------------------------------------------------------
---------------------------------------------------------------
Note
-----
- Global temporary table session private statistics used
158

back to session 1
159

---------------------------------------------------------------
---------------------------------------------------------------
---------------------------------------------------------------
160

note: default = private
161

so far ...
162

163

improved optimizer "inputs"
histograms
gather on load
private GTT
164

ISSUE #2
165

statistics needed = "∞"
166

learn ... as we go
167

learn ... from experiences
168

Adaptive Query Optimization #1
169
learn as we go

"Find all of the products with
a unit price of 15 that we have
sold more than 1 unit"
170

SQL> select ...
2 from order_iterms o,
3 product_info p
4 where o.unit_price = 15
5 and o.quantity > 1
6 and o.product_id = p.product_id
-----------------------------------------------------------------
-----------------------------------------------------------------
| 1 | NESTED LOOPS | | 12 | 200 |
| 2 | NESTED LOOPS | | 12 | 200 |
| 3 | TABLE ACCESS FULL | ORDER_ITEMS | 12 | 800 |
| 4 | INDEX UNIQUE SCAN | PRODUCT_PK | 1 | 40 |
| 5 | TABLE ACCESS BY ROWID | PRODUCT_INFO | 1 | 10 |
-----------------------------------------------------------------
171

172
what if there's 5000 ?
4 where o.unit_price = 15
5 and o.quantity > 1
-----------------------------------------------------------------
-----------------------------------------------------------------
| 1 | NESTED LOOPS | | 12 | 200 |
| 2 | NESTED LOOPS | | 12 | 200 |
| 3 | TABLE ACCESS FULL | ORDER_ITEMS | 12 | 800 |
| 4 | INDEX UNIQUE SCAN | PRODUCT_PK | 1 | 40 |
| 5 | TABLE ACCESS BY ROWID | PRODUCT_INFO | 1 | 10 |
-----------------------------------------------------------------

the runtime plan ?
173

SQL> select * from table(
2 dbms_xplan.display_cursor('...'));
--------------------------------------------------------------------------
--------------------------------------------------------------------------
| 1 | HASH JOIN | | 2000 | 16000 | 7 (0)|
| 2 | TABLE ACCESS FULL | ORDER_ITEMS | 10000 | 80000 | 7 (0)|
| 3 | TABLE ACCESS FULL | PRODUCT_INFO | 10000 | 80000 | 7 (0)|
--------------------------------------------------------------------------
174

SQL> desc V$SQL
Name Null? Type
----------------------------------------- -------- ----------------------------
SQL_TEXT VARCHAR2(1000)
SQL_FULLTEXT CLOB
SQL_ID VARCHAR2(13)
SHARABLE_MEM NUMBER
PERSISTENT_MEM NUMBER
...
IS_BIND_SENSITIVE VARCHAR2(1)
IS_BIND_AWARE VARCHAR2(1)
...
IS_REOPTIMIZABLE VARCHAR2(1)
IS_RESOLVED_ADAPTIVE_PLAN VARCHAR2(1)
...
SQL>
175

cool
176
"dodged a bullet"

Adaptive Query Optimization #2
177
learn from experience

SQL> select /*+ gather_plan_statistics */ ...
3 product_info p
4 where o.sales_date > date '2017-01-01'
5 and o.amt > 50
-------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows |
-------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 7500 | 250 |
| 1 | HASH JOIN | | 1 | 7500 | 250 |
| 2 | TABLE ACCESS FULL | ORDER_ITEMS | 1 | 7500 | 250 |
| 3 | TABLE ACCESS FULL | PRODUCT_INFO | 1 | 200 | 212 |
-------------------------------------------------------------------------
178

SQL> desc V$SQL
Name Null? Type
----------------------------------------- -------- ----------------------------
SQL_TEXT VARCHAR2(1000)
SQL_FULLTEXT CLOB
SQL_ID VARCHAR2(13)
SHARABLE_MEM NUMBER
PERSISTENT_MEM NUMBER
...
IS_BIND_SENSITIVE VARCHAR2(1)
IS_BIND_AWARE VARCHAR2(1)
...
IS_REOPTIMIZABLE VARCHAR2(1)
IS_RESOLVED_ADAPTIVE_PLAN VARCHAR2(1)
...
SQL>
179

SQL> select /*+ gather_plan_statistics */ ...
3 product_info p
5 and o.amt > 50
-------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | A-Rows |
-------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 250 | 250 |
| 1 | HASH JOIN | | 1 | 250 | 250 |
| 2 | TABLE ACCESS FULL | ORDER_ITEMS | 1 | 250 | 250 |
| 3 | TABLE ACCESS FULL | PRODUCT_INFO | 1 | 212 | 212 |
-------------------------------------------------------------------------
Note:
- statistics feedback used for this statement
180

seems cool
181

we mightta got carried away :-(
182

SQL> select
2 directive_id,
3 type,
4 reason
5 from DBA_SQL_PLAN_DIRECTIVES;
DIRECTIVE_ID TYPE REASON
---------------------- ----------------------- ------------------------------------
14774885086507357845 DYNAMIC_SAMPLING JOIN CARDINALITY MISESTIMATE
17607835686132426105 DYNAMIC_SAMPLING SINGLE TABLE CARDINALITY MISESTIMATE
28719982731298412 DYNAMIC_SAMPLING_RESULT SINGLE TABLE CARDINALITY MISESTIMATE
7592221435558555884 DYNAMIC_SAMPLING SINGLE TABLE CARDINALITY MISESTIMATE
...
...
183

SQL> select ...
3 product_info p
Note:
- 3 Sql Plan Directives used for this statement
184

same SQL ...
185

... different plans
186

... then more different plans
187

... then even more different plans
188

So ... this happens
189

upgrade to 12.1
190

new optimizations
191

new statistics
192

default = everything enabled
193

"the plans keep changing"
194

just one switch
195

196

options
197

optimizer_adaptive_features = false
198

optimizer_features_enable = 11.2.0.4
199

don't panic ... that's fine
endorsed
200

12.1 ... perhaps too aggressive
201

12.2 ... more control
202

12.2 ... more passive
203

TRUE FALSE 204

12.2 ... much better
205

12.1
patch 22652097
206

which brings us to ...
207

upgrade scenarios
208

there are only two
209

1) your system sucks :-)
210
upgrade to 12.2
explore full adaptive

2) your system is OK
211
how can I lower risk ?

SQL Plan Baselines
212

apology #1
213

214

but ultimately ...
215

... just three things
216

try run stuff good
1
217

when it's running good …
2
218

… keep running it like that !
219

If there is something better
220
3

then tell me
221

222

223

baselines do exactly this !
224

rarely used
225

apology #2
226
evolve
accepted
repeatable
unaccepted
capture
signature

close to our goals
227
working good ? Keep it
found something better ? Tell me
found something worse ? Don’t use it

not new
228
11g

(quick) terminology
229

signature
230

SQL> select sal
from emp
where empno = ...
SQL> select SAL
from EMP where EMPNO = ...
SQL> SELECT /* comment */ SAL FROM EMP
WHERE EMPNO = ...
etc
etc
231
signature

plan(s) for signature
232

233
SELECT A.DATA,
B.DATA
FROM TAB1 A, TAB2
WHERE B.TAB1_ID = A.ID
AND A.CODE = :1
-----------------------------------------------------------------------
-----------------------------------------------------------------------
| 0 | SELECT STATEMENT | | | |
| 1 | NESTED LOOPS | | 25 | 425 |
| 2 | NESTED LOOPS | | 25 | 425 |
|* 3 | TABLE ACCESS FULL | TAB1 | 1 | 11 |
|* 4 | INDEX RANGE SCAN | TAB2_TAB1_FKI | 25 | |
| 5 | TABLE ACCESS BY INDEX ROWID| TAB2 | 25 | 150 |
-----------------------------------------------------------------------
select a.data, b.data
from tab1 a, tab2
where b.tab1_id = a.id
and a.code = :1
"capture" a baseline

234
SELECT A.DATA,
B.DATA
FROM TAB1 A, TAB2
AND A.CODE = :1
-----------------------------------------------------------------------
-----------------------------------------------------------------------
| 1 | NESTED LOOPS | | 25 | 425 |
| 2 | NESTED LOOPS | | 25 | 425 |
-----------------------------------------------------------------------
------------------------------------------------------------------------
------------------------------------------------------------------------
| * 1 | HASH JOIN | | 25 | 425 |
| * 5 | TABLE ACCESS FULL | TAB1 | 1 | 11 |
| 8 | TABLE ACCESS FULL | TAB2 | 25 | 150 |
------------------------------------------------------------------------
from tab1 a, tab2
and a.code = :1

when it's running good …
235
… keep running it like that !

236
SELECT A.DATA,
B.DATA
FROM TAB1 A, TAB2
AND A.CODE = :1
-----------------------------------------------------------------------
-----------------------------------------------------------------------
| 1 | NESTED LOOPS | | 25 | 425 |
| 2 | NESTED LOOPS | | 25 | 425 |
-----------------------------------------------------------------------
------------------------------------------------------------------------
------------------------------------------------------------------------
| * 1 | HASH JOIN | | 25 | 425 |
------------------------------------------------------------------------
from tab1 a, tab2
and a.code = :1
but we saved this !

maybe it is better ?
237
------------------------------------------------------------------------
------------------------------------------------------------------------
| * 1 | HASH JOIN | | 25 | 425 |
------------------------------------------------------------------------

evolve a plan
238

DBMS_SPM.EVOLVE_SQL_PLAN_BASELINE
239

240
SELECT A.DATA,
B.DATA
FROM TAB1 A, TAB2
AND A.CODE = :1
-----------------------------------------------------------------------
-----------------------------------------------------------------------
| 1 | NESTED LOOPS | | 25 | 425 |
| 2 | NESTED LOOPS | | 25 | 425 |
-----------------------------------------------------------------------
------------------------------------------------------------------------
------------------------------------------------------------------------
| * 1 | HASH JOIN | | 25 | 425 |
------------------------------------------------------------------------
from tab1 a, tab2
and a.code = :1

unless …
241
… that makes it bad or
there's something better

if only we had done ...
242
DBMS_SPM.I_LOVE_THIS_PLAN(sql_id)
DBMS_SPM. THIS_PLAN_SUCKS(sql_id)
DBMS_SPM. SWAP_THIS_PLAN_IN(sql_id)

so back to upgrade
243

2) your system is OK
244
how can I lower risk ?

SQL> alter system set
2 optimizer_capture_sql_plan_baselines = true
245

SQL> exec DBMS_SPM.PACK_STGTAB_BASELINE
(datapump)
SQL> exec DBMS_SPM.UNPACK_STGTAB_BASELINE
246

the same performance
247

248

wrap up
249

12c optimizer is better
250

12c optimizer is different
251

12.2+ preferred
252

12.1 ... backport
253

SPM is your friend
254

Thank you!
youtube bit.ly/youtube-connor
blog bit.ly/blog-connor
twitter bit.ly/twitter-connor
400+ posts mainly on database & development
250 technical videos, new uploads every week
rants and raves on tech and the world :-)

Melbourne Groundbreakers Tour - Upgrading without risk

More Related Content

What's hot (20)

Similar to Melbourne Groundbreakers Tour - Upgrading without risk (20)

More from Connor McDonald (20)

Recently uploaded (20)

Melbourne Groundbreakers Tour - Upgrading without risk

Editor's Notes