-- Copyright 2018 Tanel Poder. All rights reserved. More info at http://tanelpoder.com

-- Licensed under the Apache License, Version 2.0. See LICENSE.txt for terms & conditions.

--------------------------------------------------------------------------------

--

-- File name: demos/bind_peeking_nested_loops.sql

-- Purpose: this script demos how a "wrong" bind variable value

-- can cause an execution plan to be compiled which is

-- very inefficient for the next execution with different bind variable

-- values (with large number of matching rows)

-- the second execution of the query takes very long time to complete

-- despite adaptive bind variable peeking, which would kick in during the

-- next (3rd) execution

--

-- This problem happens even on Oracle 11.2 despite adaptive bind peeking

-- and cardinality feedback (due to design, not a bug)

--

-- Author: Tanel Poder (tanel@e2sn.com)

-- Copyright: (c) http://tech.e2sn.com

--

--------------------------------------------------------------------------------

alter session set optimizer_use_sql_plan_baselines = false;

DROP TABLE t_bp1;

DROP TABLE t_bp2;

DROP TABLE t_bp3;

DROP TABLE t_bp4;

DROP TABLE t_bp5;

CREATE TABLE t_bp1 AS SELECT * FROM dba_objects WHERE rownum <= 50000;

CREATE TABLE t_bp2 AS SELECT * FROM dba_objects WHERE rownum <= 10000;

CREATE TABLE t_bp3 AS SELECT * FROM dba_objects WHERE rownum <= 10000;

CREATE TABLE t_bp4 AS SELECT * FROM dba_objects WHERE rownum <= 10000;

CREATE TABLE t_bp5 AS SELECT * FROM dba_objects WHERE rownum <= 10000;

CREATE INDEX i_bp1 ON t_bp1(owner);

CREATE INDEX i_bp2 ON t_bp2(owner);

CREATE INDEX i_bp3 ON t_bp3(owner);

CREATE INDEX i_bp4 ON t_bp4(owner);

CREATE INDEX i_bp5 ON t_bp5(owner);

EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T_BP1',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 254');

EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T_BP2',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 254');

EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T_BP3',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 254');

EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T_BP4',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 254');

EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T_BP5',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 254');

-- EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T1',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 1');

-- EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T2',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 1');

-- EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T3',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 1');

-- EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T4',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 1');

-- EXEC DBMS_STATS.GATHER_TABLE_STATS(user,'T5',cascade=>TRUE, method_opt=>'FOR ALL INDEXED COLUMNS SIZE 1');

-- this hack might be needed to "help" this problem to show up sometimes:

-- ALTER SESSION SET OPTIMIZER_INDEX_COST_ADJ=10;

CREATE OR REPLACE PROCEDURE test_bp (num_loops IN NUMBER, sleep IN NUMBER

, v1 IN VARCHAR2, v2 IN VARCHAR2, v3 IN VARCHAR2, v4 IN VARCHAR2, v5 IN VARCHAR2)

AS

r1 DATE;

r2 DATE;

s NUMBER;

BEGIN

FOR i IN 1..num_loops LOOP

SELECT /*+ opt_param('_optimizer_use_feedback', 'false') */

MIN(t_bp1.created), MAX(t_bp5.created) INTO r1, r2

FROM

t_bp1

, t_bp2

, t_bp3

, t_bp4

, t_bp5

WHERE

t_bp1.object_id = t_bp2.object_id

AND t_bp2.object_id = t_bp3.object_id

AND t_bp3.object_id = t_bp4.object_id

AND t_bp4.object_id = t_bp5.object_id

AND t_bp1.owner = v1

AND t_bp2.owner = v2

AND t_bp3.owner = v3

AND t_bp4.owner = v4

AND t_bp5.owner = v5;

s := s + (r2 - r1); -- dummy calculation

IF sleep > 0 THEN DBMS_LOCK.SLEEP(sleep); END IF;

END LOOP;

DBMS_OUTPUT.PUT_LINE(TO_CHAR(s));

END;

/

PROMPT Running query first time, this should be fast (and should use nested loops execution plan)

--SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY_CURSOR(null,null));

SET TIMING ON

PROMPT Now running the same query with different bind variables (this query should take very long time)