SAGE Computing Services

Customised Oracle Training Workshops and Consulting

Bulk Binds
Be A Bulk Binding Baron

Scott Wesley
Systems Consultant
& Trainer







who_am_i;
http://strategy2c.wordpress.com/2009/01/10/strategy-for-goldfish-funny-
illustration-by-frits/
bulk binds
been around since 8i
not used often enough
binding is the assignment of values to
placeholders in SQL statements
my_column = :my_value
Bulk Binding is binding an array of values
my_column = my_array(i)
in-bind
my_variable
insert
update
merge
out-bind
my_variable
returning
define
my_variable
fetch
select
multiset
A quick word on Collections
lists or sets of information
invaluable
can be faster than SQL
cache frequent access in PGA
my_module (pt_array IN OUT NOCOPY t_array)
types of collections
Associative Arrays (aka index-by tables)
Only PL/SQL
Sparse
Can have negative indexes
Similar to hash table concept
Different datatypes for indexes
TYPE type_name IS TABLE OF element_type [NOT NULL]
INDEX BY [PLS_INTEGER | BINARY_INTEGER |
VARCHAR2(size_limit)];

TYPE emp_tab IS TABLE OF
employees%ROWTYPE
INDEX BY PLS_INTEGER;

Nested Tables
Can use in SQL
Unbounded
Good for set operations
TYPE type_name IS TABLE OF element_type [NOT NULL];

TYPE top_emp IS TABLE OF
employees.employee_id%TYPE;

CREATE TYPE galaxy AS TABLE OF
star_object;

VARRAY
Specify maximums
Dense
Can use in SQL
Most similar to 3GL array
TYPE type_name IS {VARRAY | VARYING ARRAY} (size_limit)
OF element_type [NOT NULL];

TYPE Calendar IS VARRAY(366) OF DATE;

CREATE TYPE rainbow AS VARRAY(7) OF VARCHAR2(64);

CREATE TYPE solar_system AS VARRAY(8) OF
planet_object;

back to the business of binds
from the 8.1.7 documentation
Conventional Bind
Performance penalty for many context switches
Oracle Server
PL/SQL Runtime Engine
PL/SQL block
Procedural
statement
executor
SQL Engine
SQL statement
executor
FOR r_emp IN c_emp LOOP
UPDATE emp
SET sal = sal * 1.1
WHERE empno = r_emp.empno;
END LOOP;
Bulk Bind
Much less overhead
Oracle Server
PL/SQL Runtime Engine
PL/SQL block
Procedural
statement
executor
SQL Engine
SQL statement
executor
FORALL i IN INDICES OF t_emp
UPDATE emp
SET sal = sal * 1.1
WHERE empno = t_emp(i);

when would you use it?
plsql_optimize_level = 42
lifesgoldenrule.com
no bulk bind do it in SQL
If the DML statement affects four or
more database rows, the use of bulk SQL
can improve performance considerably
- Oracle Documentation
SELECT everyone
BULK COLLECT
INTO my_array
FROM the_worlds_idiots;
SELECT everyone
BULK COLLECT
INTO my_array LIMIT 100
FROM the_worlds_idiots;
reoccurring SQL within PL/SQL loop
DML LOOP FORALL
multiple operations on dataset BULK COLLECT
http://www.sharenator.com/Before_After/stinejensen_530-65109.html
11g compound triggers auditing
finally some coding examples?
DML Bind Example
-- Slower method, running the UPDATE
-- statements within a regular loop
FOR i IN t_emp.FIRST..t_emp.LAST LOOP
UPDATE emp_largish
SET sal = sal * 1.1
WHERE empno = t_emp(i);
END LOOP;

1.94 seconds

-- Efficient method, using a bulk bind on a VARRAY
FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp_largish
SET sal = sal * 1.1
WHERE empno = t_emp(i);

1.40 seconds
Collect Example
-- Slower method, assigning each
-- collection element within a loop.
v_index := 0;
FOR r_emp IN c_emp LOOP
v_index := v_index + 1;
t_emp(v_index).empno := r_emp.empno;
t_emp(v_index).ename := r_emp.ename;
END LOOP;

0.19 seconds

-- Efficient method, using a bulk bind
OPEN c_emp;
FETCH c_emp BULK COLLECT INTO t_emp;
CLOSE c_emp;

0.09 seconds

Example of Combination
-- Slower method, running update
-- statement within regular loop,
-- returning individual elements
FOR i IN t_emp.FIRST..t_emp.LAST LOOP
UPDATE emp_largish
SET sal = sal * 1.1
WHERE empno = t_emp(i)
RETURNING sal
INTO t_sal(i);
END LOOP;

0.75 seconds

-- Efficient method, using a bulk bind
-- and bulk collect
FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp_largish
SET sal = sal * 1.1
WHERE empno = t_emp(i)
RETURNING sal BULK COLLECT
INTO t_sal;

0.29 seconds
when wouldnt you use it
lifesgoldenrule.com
can you just do it in SQL?
when you only need one iteration
just filter records with
where/intersect/minus
when you haven't tested it
with & without
>= 10g may silently do it for you
too much data?
what if I have a sparse collection?
Uh oh
DECLARE
TYPE emp_tab IS TABLE OF emp.empno%TYPE INDEX BY PLS_INTEGER;
t_emp emp_tab;
CURSOR c_emp IS
SELECT empno FROM emp;
BEGIN
OPEN c_emp;
FETCH c_emp BULK COLLECT INTO t_emp;
CLOSE c_emp;

-- ...do a bunch of processing, including
-- something like:
t_emp.DELETE(3);

FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp
SET sal = sal * 1.1
WHERE empno = t_emp(i);
END;
/

ORA-22160: element at index [3] does not exist
ORA-06512: at line 15
processing spare collections
introduced in 10g
INDICES OF
allows iteration of all index values,
not just from lower to upper bound
(tab.FIRST..tab.LAST)
VALUES OF
Create PL/SQL table effectively
indexing your collection
- process specific elements
- process in particular order
- process an element more than once
simple
FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp
SET sal = sal * 1.1
WHERE empno = t_emp(i);

FIRST
LAST
Updated all rows
In order of array listing
simple
FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp
SET sal = sal * 1.1
WHERE empno = t_emp(i);

FIRST
LAST
ORA-22160: element at index [3] does not exist
clever
FORALL i IN INDICES OF t_emp
UPDATE emp
SET comm = comm * 1.5
WHERE empno = t_emp(i);

7890
7988
7752
7521
7900
Updated rows within array
In order of array listing
t_emp(3) := 7890
t_emp(5) := 7988
t_emp(6) := 7752
t_emp(8) := 7521
t_emp(9) := 7900
advanced
7890
7988
7752
t_index(-2):= 6
t_index( 0):= 3
t_index( 3):= 5
t_index(10):= 6
t_emp(9) := 7900
t_emp(8) := 7521
t_emp(6) := 7752
t_emp(5) := 7988
t_emp(3) := 7890
FORALL i IN VALUES OF t_index
UPDATE emp
SET comm = comm * 1.5
WHERE empno = t_emp(i);

DECLARE
TYPE emp_tab IS TABLE OF emp.empno%TYPE INDEX BY PLS_INTEGER;
TYPE numlist IS TABLE OF PLS_INTEGER INDEX BY BINARY_INTEGER;
t_emp emp_tab;
t_indexes numlist;
BEGIN
t_emp(3) := 7890;
t_emp(5) := 7988;
t_emp(6) := 7752;
t_emp(8) := 7521;
t_emp(9) := 7900;

t_indexes(-2) := 6;
t_indexes(0) := 3;
t_indexes(3) := 5;
t_indexes(10) := 6;

what if I have 9i?
CREATE OR REPLACE PACKAGE BODY sw_bulk_insert IS
PROCEDURE sw_insert (sw_tab IN t_sw_tab) IS
BEGIN
$IF dbms_db_version.ver_le_9 $THEN
DECLARE
l_dense t_sw_tab;
ln_index PLS_INTEGER := sw_tab.FIRST;
BEGIN
<< dense_loop >>
WHILE (l_index IS NOT NULL) LOOP
l_dense(l_dense.COUNT + 1) := sw_tab(l_index);
l_index := sw_tab.NEXT(l_index);
END LOOP dense_loop;

FORALL i IN 1..l_dense.COUNT
INSERT INTO sw_table VALUES l_dense(i);
END;
$ELSE
FORALL i IN INDICES OF sw_tab
INSERT INTO sw_table
VALUES sw_tab(i);
$END
END sw_insert;
END sw_bulk_insert;

DECLARE
t_emp t_emp_obj;
BEGIN
SELECT emp_obj(empno, ename, job, mgr, hiredate, sal,
comm, deptno)
BULK COLLECT INTO t_emp
FROM emp;

-- Remove those with commission to create sparse
collection
FOR i IN 1..t_emp.COUNT LOOP
IF t_emp(i).comm IS NOT NULL THEN
t_emp.DELETE(i);
END IF;
END LOOP;

-- No need for FORALL
INSERT INTO emp2
SELECT * FROM TABLE(CAST(t_emp AS t_emp_obj));
END;
/

CREATE TABLE emp2 AS SELECT * FROM emp WHERE 1=0;
CREATE OR REPLACE TYPE emp_obj AS OBJECT
(empno NUMBER(4)
,ename VARCHAR2(10)
,job VARCHAR2(9)
,mgr NUMBER(4)
,hiredate DATE
,sal NUMBER(7,2)
,comm NUMBER(7,2)
,deptno NUMBER(2))
/

CREATE OR REPLACE TYPE t_emp_obj AS TABLE OF emp_obj
/

can I do this "bunch of processing" in bulk, outside the db?
DECLARE
TYPE emp_tab IS TABLE OF emp.empno%TYPE INDEX BY PLS_INTEGER;
t_emp emp_tab;
CURSOR c_emp IS
SELECT empno FROM emp;
BEGIN
OPEN c_emp;
FETCH c_emp BULK COLLECT INTO t_emp;
CLOSE c_emp;

-- ...do a bunch of processing, including
-- something like:
t_emp.DELETE(3);

FORALL i IN t_emp.FIRST..t_emp.LAST
UPDATE emp
SET sal = sal * 1.1
WHERE empno = t_emp(i);
END;
/

-- ... Or just those employees that exist in another collection
-- t_emp intersect with t_emp_leaders;

MULTISET operations
nested tables ONLY
DECLARE
TYPE colours IS TABLE OF VARCHAR2(64);
comp1 colours;
comp2 colours;
result colours;

PROCEDURE show_table (p_table IN colours) IS
BEGIN
FOR i IN p_table.FIRST..p_table.LAST LOOP
DBMS_OUTPUT.PUT (p_table(i)||' ');
END LOOP;
DBMS_OUTPUT.NEW_LINE;
END;
BEGIN
comp1 := colours('Black','White','Red','Red');
comp2 := colours('Black','Red','Yellow');

result := comp1 MULTISET UNION comp2;
dbms_output.put_line ('multiset union is ');
show_table(result);

END;
/

multiset union is
Black White Red Red Black Red Yellow
...
BEGIN
comp1 := colours('Black','White','Red','Red');
comp2 := colours('Black','Red','Yellow');

result := comp1 MULTISET UNION DISTINCT comp2;
dbms_output.put_line ('multiset union distinct is ');
show_table(result);

END;
/

multiset union distinct is
Black White Red Yellow
...
BEGIN
comp1 := colours('Black','White','Red','Red');
comp2 := colours('Black','Red','Yellow');

result := comp1 MULTISET INTERSECT DISTINCT comp2;
dbms_output.put_line ('multiset intersect distinct is ');
show_table(result);

END;
/

multiset intersect distinct is
Black Red
...
BEGIN
comp1 := colours('Black','White','Red','Red');
comp2 := colours('Black','Red','Yellow');

result := comp1 MULTISET EXCEPT DISTINCT comp2;
dbms_output.put_line ('multiset except distinct is ');
show_table(result);

END;
/

multiset except distinct is
White
...
BEGIN
comp1 := colours('Black','White','Red','Red');

comp1 := comp1 MULTISET UNION DISTINCT comp1;
dbms_output.put_line (self multiset intersect distinct is ');
show_table(comp1);

END;
/

self multiset union distinct is
Black White Red
DECLARE
TYPE colours IS TABLE OF VARCHAR2(64);
t_colours colours := colours('Black','White','Red','Red');

PROCEDURE show_table (p_table IN colours)
...
BEGIN
DBMS_OUTPUT.PUT_LINE('Count:'||CARDINALITY(t_colours));
DBMS_OUTPUT.PUT_LINE('Distinct:'||CARDINALITY(SET(t_colours)));
IF t_colours IS NOT A SET THEN
t_colours := SET(t_colours);
show_table(t_colours);
END IF;
END;
/

Count:4
Distinct:3
Black White Red
but wait, there's still more...
Equality
DECLARE
TYPE colours IS TABLE OF VARCHAR2 (64);
group1 colours := colours ('Black', 'White');
group2 colours := colours ('White', 'Black');
group3 colours := colours ('White', 'Red');
BEGIN

IF group1 = group2 THEN
DBMS_OUTPUT.PUT_LINE ('Group 1 = Group 2');
ELSE
DBMS_OUTPUT.PUT_LINE ('Group 1 != Group 2');
END IF;


IF group2 != group3 THEN
DBMS_OUTPUT.PUT_LINE ('Group 2 != Group 3');
ELSE
DBMS_OUTPUT.PUT_LINE ('Group 2 = Group 3');
END IF;
END;
/
DECLARE
TYPE colours IS TABLE OF VARCHAR2(64);
t_colours colours := colours('Black','White','Red','Red');
v_colour VARCHAR2(64) := 'Black';
BEGIN
IF v_colour MEMBER OF t_colours THEN
DBMS_OUTPUT.PUT_LINE('Exists');
ELSE
DBMS_OUTPUT.PUT_LINE(Absent');
END IF;
END;
/

Exists
Membership
DECLARE
TYPE colours IS TABLE OF VARCHAR2(64);
t_colours colours := colours('Black','White','Red','Yellow');
t_colours2 colours := colours('Black','Red');
t_colours3 colours := colours('Black','Blue');
BEGIN
IF t_colours2 SUBMULTISET OF t_colours THEN
DBMS_OUTPUT.PUT_LINE('2 Subset');
ELSE
DBMS_OUTPUT.PUT_LINE('2 Separate');
END IF;

IF t_colours3 SUBMULTISET OF t_colours THEN
DBMS_OUTPUT.PUT_LINE('3 Subset');
ELSE
DBMS_OUTPUT.PUT_LINE('3 Separate');
END IF;
END;
/

2 Subset
3 Separate
Subset
DECLARE
TYPE t_c IS TABLE OF VARCHAR2(32767) INDEX BY
PLS_INTEGER;
l_c t_c;
BEGIN
SELECT LPAD('x',32767,'x') bb
BULK COLLECT
INTO l_c
FROM DUAL
CONNECT BY LEVEL <=1000000;
END;
/
declare
*
ERROR at line 1:
ORA-04030: out of process memory when trying to allocate
16396 bytes (koh-kghu call ,pl/sql vc2)
ORA-06512: at line 5

Elapsed: 00:00:02.70
Boom!
limiting data
prevent collections from expanding with no limit
SELECT sal BULK COLLECT
INTO t_emp
FROM emp
WHERE ROWNUM <= 10;

WHERE ROWNUM BETWEEN 11 AND 20 Wont work


FROM emp SAMPLE(10) -- Random

-- Performance hit:
SELECT * FROM (
SELECT ROW_NUMBER() OVER (ORDER BY empno)
AS from_to_rank
)
WHERE from_to_rank BETWEEN 1 AND 10

LIMIT clause
DECLARE
v_rows PLS_INTEGER := 10;
BEGIN
OPEN c_emp;
LOOP
FETCH c_emp BULK COLLECT INTO t_emp LIMIT v_rows;

EXIT WHEN t_emp.COUNT = 0;
null; -- (Process information)

END LOOP;
CLOSE c_emp;
take care with the exit condition
OPEN c_emp;
LOOP
FETCH c_emp BULK COLLECT INTO t_emp LIMIT v_rows;

EXIT WHEN t_emp.COUNT = 0; -- Here?
null; -- (Process information)
EXIT WHEN c_emp%NOTFOUND; -- or here?

END LOOP;
CLOSE c_emp;
EXCEPTION WHEN NO_DATA_FOUND
-- Does this get raised?
t_emp.COUNT = 0
** Count:10
CLARK
JAMES
MARTIN
MILLER
WARD
KING
ALLEN
ADAMS
SMITH
JONES
** Count:4
TURNER
BLAKE
SCOTT
FORD
** Count:0

PL/SQL procedure successfully completed.
LOOP
FETCH c_emp BULK COLLECT INTO t_emp LIMIT 10;
dbms_output.put_line('** Count:'||t_emp.COUNT);

EXIT WHEN t_emp.COUNT = 0;
dbms_output.put_line(t_emp(i).ename);

-- EXIT WHEN c_emp%NOTFOUND;

WHEN c_emp%NOTFOUND
** Count:10
CLARK
JAMES
MARTIN
MILLER
WARD
KING
ALLEN
ADAMS
SMITH
JONES
** Count:4
TURNER
BLAKE
SCOTT
FORD

PL/SQL procedure successfully completed.
LOOP
FETCH c_emp BULK COLLECT INTO t_emp LIMIT 10;
dbms_output.put_line('** Count:'||t_emp.COUNT);

-- EXIT WHEN t_emp.COUNT = 0;
dbms_output.put_line(t_emp(i).ename);

EXIT WHEN c_emp%NOTFOUND;

implicit cursors dont bulk up
DECLARE
TYPE emp_tab IS TABLE OF emp.empno%TYPE INDEX BY
PLS_INTEGER;
t_emp emp_tab;
BEGIN
SELECT empno
BULK COLLECT INTO t_emp
FROM emp
LIMIT 10;
END;
/
LIMIT 10
*
ERROR at line 8:
ORA-06550: line 8, column 9:
PL/SQL: ORA-00933: SQL command not properly ended
ORA-06550: line 5, column 3:
PL/SQL: SQL Statement ignored
and if I have no data?
NO_DATA_FOUND
DECLARE
v_sal emp.sal%TYPE;
BEGIN
SELECT sal
INTO v_sal
FROM emp
WHERE 1=0;
END;
/
DECLARE
*
ERROR at line 1:
ORA-01403: no data found
ORA-06512: at line 4

DECLARE
TYPE emp_tab IS TABLE OF emp.sal%TYPE
INDEX BY PLS_INTEGER;
t_emp emp_tab;
BEGIN
SELECT sal BULK COLLECT
INTO t_emp
FROM emp
WHERE 1=0;
END;
/
PL/SQL procedure successfully completed.
bulk attributes
Bulk Cursor Attributes
DECLARE
TYPE dept_list IS TABLE OF NUMBER;
t_dept dept_list := dept_list(10, 20, 30);
BEGIN
FORALL i IN t_dept.FIRST..t_dept.LAST
DELETE FROM emp WHERE deptno = t_dept(i);
-- Each indice may update multiple rows
dbms_output.put_line('Total rows affected: '||SQL%ROWCOUNT);

-- How many rows were affected by each delete statement?
FOR j IN t_dept.FIRST..t_dept.LAST LOOP
dbms_output.put_line('Dept '||t_dept(j)||
'rows:'||SQL%BULK_ROWCOUNT(j));
END LOOP;
END;
/

Total rows affected: 14
Dept 10 rows:3
Dept 20 rows:5
Dept 30 rows:6

PL/SQL procedure successfully completed.
what if we have exceptions for a particular row?
CREATE TABLE emp2 AS SELECT * FROM emp;
ALTER TABLE emp2 ADD CONSTRAINT be_test CHECK (sal = ROUND(sal));

DECLARE
TYPE emp_list IS TABLE OF NUMBER;
t_emp emp_list := emp_list(7934, 7782, 7839, 7788, 7900);
e_dml_errors EXCEPTION;
PRAGMA exception_init(e_dml_errors, -24381);

BEGIN
FORALL i IN t_emp.FIRST..t_emp.LAST SAVE EXCEPTIONS
-- Will raise exception for sal of 1250 but not 1300
UPDATE emp2
SET sal = sal/100
WHERE empno = t_emp(i);
dbms_output.put_line('Total rows affected:
||SQL%ROWCOUNT);
EXCEPTION
WHEN e_dml_errors THEN -- Now we figure out what failed and why.
dbms_output.put_line('Total failed updates: '||SQL%BULK_EXCEPTIONS.COUNT);
FOR j IN 1..SQL%BULK_EXCEPTIONS.COUNT LOOP
dbms_output.put_line('Error '||j||' for indice
||SQL%BULK_EXCEPTIONS(j).ERROR_INDEX);
dbms_output.put_line('Error message is
||SQLERRM(-SQL%BULK_EXCEPTIONS(j).ERROR_CODE));
END LOOP;
END;

Total failed updates: 2
Error 1 for indice 2
Error message is ORA-02290: check constraint (.) violated
Error 2 for indice 5
Error message is ORA-02290: check constraint (.) violated

PL/SQL procedure successfully completed.
Bulk Exceptions
select empno, ename, sal from emp2 where
empno in (7934, 7782, 7839, 7788, 7900);

EMPNO ENAME SAL
---------- ---------- ----------
7934 MILLER 13
7782 CLARK 2450
7839 KING 50
7788 SCOTT 30
7900 JAMES 950

5 rows selected.
41%
33%
10%
9%
7%
FOR
CURSOR
COLLECT
LIMIT100
IISS
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
10g 11g
Indices1000
Indices100
Limit1000
Limit100
0
50
100
150
200
250
300
9i 10g 11g
Limit 100
Limit 1000
0%
20%
40%
60%
80%
100%
9
i
1
0
g
1
1
g
IISS
Limit100
Collect
Cursor
For
FOR CURSOR COLLECT LIMIT100 IISS
Enterprise Manager
Limit 100 vs 1000
Separate runs of
For | Cursor | Collect | Limit | IISS
Whats new?
8i
Bulk operations!
Collections indexed by BINARY_INTEGER only

9i
Bulk collect via dynamically opened ref cursor

9i Release 2
Bulk collect into collection of records
Index by VARCHAR2 & BINARY_INTEGER (or their subtypes)

10g
PL/SQL engine may decide to peform array fetching for you
INDICES OF/VALUES OF introduced
MULTISET operations

11g
table(bulk_index).field is now supported at run-time
DBMS_SQL allows bulk binds using user-define collection types


CREATE TABLE emp2 AS SELECT empno,ename,sal, SYSTIMESTAMP ts
FROM emp WHERE 1=0;
DECLARE
CURSOR sw IS SELECT empno, ename, sal FROM emp;
TYPE t_sw IS TABLE OF sw%ROWTYPE;
lt_sw t_sw;
BEGIN
OPEN sw;
FETCH sw BULK COLLECT INTO lt_sw;
CLOSE sw;
FORALL i IN lt_sw.FIRST..lt_sw.LAST
INSERT INTO emp2
VALUES (lt_sw(i).empno
,lt_sw(i).ename
,lt_sw(i).sal
,SYSTIMESTAMP);
END;
/

values (lt_sw(i).empno
*
ERROR at line 11:
ORA-06550: line 11, column 9:
PLS-00436: implementation restriction: cannot reference fields of
BULK In-BIND table of records
some final words
Read
appropriate
documentation
READ THIS TWICE!
Tom Kyte
In search of the truth - Or Correlation is not Causation
http://asktom.oracle.com/pls/ask/download_file?p_file=30671718
13508366601

If any paper you read tells you what but not why, regardless of
its timestamp, I suggest you take its advice with a big grain of
salt as it may be applicable only to only specific cases and
not applicable to you at all. In fact, it could be something that
was true in the past and not true anymore. If they give you no
way to find out, ignore it.

things change
it only takes one case to show something is...
not universally true
Rules of thumb without evidence, without ways to test them before
implementing them, are dangerous, seriously dangerous
Because a rule of thumb is a principle with
broad application not intended to be strictly
accurate or reliable for all situations

SAGE Computing Services
Customised Oracle Training Workshops and Consulting






Questions and Answers?





Presentations are available from our website:
http://www.sagecomputing.com.au

[email protected]
[email protected]
http://triangle-circle-square.blogspot.com