簡(jiǎn)單試驗(yàn)一下Bulk Binds對(duì)性能的提高
http://www.itpub.net/130636.html
當(dāng)Oracle運(yùn)行PL/SQL時(shí)會(huì)使用兩套引擎，所有procedural code由PL/SQL engine 完成，所有SQL由SQLengine處理。所以如果Oracle從一個(gè)collection中循環(huán)執(zhí)行相同的DML操作，那么為了避免兩套engine切換所消耗的系統(tǒng)資源，可以使用bulk binds來(lái)把所有的DML操作binding到一次操作中完成。這將極大提高PL/SQL的執(zhí)行效率。
以下是簡(jiǎn)單的測(cè)試，用兩種方式插入100000條數(shù)據(jù)，可以看到效率提高了7倍左右。
PHP code:
SQL> CREATE TABLE test1(
2 id NUMBER(10),
3 description VARCHAR2(50));
Table created
SQL> ALTER TABLE test1 ADD (
2 CONSTRAINT test1_pk PRIMARY KEY (id));
Table altered
SQL> SET TIMING ON;
SQL> DECLARE
2 TYPE id_type IS TABLE OF test1.id%TYPE;
3 TYPE description_type IS TABLE OF test1.description%TYPE;
4
5 t_id id_type := id_type();
6 t_description description_type := description_type();
7 BEGIN
8 FOR i IN 1 .. 100000 LOOP
9 t_id.extend;
10 t_description.extend;
11
12 t_id(t_id.last) := i;
13 t_description(t_description.last) := 'Description: ' || To_Char(i);
14 END LOOP;
15
16 FOR i IN t_id.first .. t_id.last LOOP
17 INSERT INTO test1 (id, description)
18 VALUES (t_id(i), t_description(i));
19 END LOOP;
20
21 COMMIT;
22 END;
23 /
PL/SQL procedure successfully completed
Executed in 141.233 seconds
SQL> truncate table test1;
Table truncated
Executed in 0.631 seconds
SQL>
SQL> DECLARE
2 TYPE id_type IS TABLE OF test1.id%TYPE;
3 TYPE description_type IS TABLE OF test1.description%TYPE;
4
5 t_id id_type := id_type();
6 t_description description_type := description_type();
7 BEGIN
8 FOR i IN 1 .. 100000 LOOP
9 t_id.extend;
10 t_description.extend;
11
12 t_id(t_id.last) := i;
13 t_description(t_description.last) := 'Description: ' || To_Char(i);
14 END LOOP;
15
16 FORALL i IN t_id.first .. t_id.last
17 INSERT INTO test1 (id, description)
18 VALUES (t_id(i), t_description(i));
19
20 COMMIT;
21 END;
22 /
PL/SQL procedure successfully completed
Executed in 27.52 seconds
SQL> select count(*) from test1;
COUNT(*)
----------
100000
Executed in 0.04 seconds
SQL>
下面我們使用上面那個(gè)例子中插入的100000條數(shù)據(jù)，來(lái)測(cè)試一下BULK COLLECT的威力。PHP code:
SQL> SET TIMING ON;
SQL>
SQL> DECLARE
2 TYPE id_type IS TABLE OF test1.id%TYPE;
3 TYPE description_type IS TABLE OF test1.description%TYPE;
4
5 t_id id_type := id_type();
6 t_description description_type := description_type();
7
8 CURSOR c_data IS
9 SELECT *
10 FROM test1;
11 BEGIN
12 FOR cur_rec IN c_data LOOP
13 t_id.extend;
14 t_description.extend;
15
16 t_id(t_id.last) := cur_rec.id;
17 t_description(t_description.last) := cur_rec.description;
18 END LOOP;
19 END;
20 /
PL/SQL procedure successfully completed
Executed in 2.974 seconds
SQL>
SQL> DECLARE
2 TYPE id_type IS TABLE OF test1.id%TYPE;
3 TYPE description_type IS TABLE OF test1.description%TYPE;
4
5 t_id id_type;
6 t_description description_type;
7 BEGIN
8 SELECT id, description
9 BULK COLLECT INTO t_id, t_description FROM test1;
10 END;
11 /
PL/SQL procedure successfully completed
Executed in 0.371 seconds
SQL>
結(jié)論：當(dāng)我們需要將大量的檢索結(jié)果放入一個(gè)collection的時(shí)候，使用bulking將比直接使用cursor循環(huán)有效的多。
關(guān)于BulkBinds中LIMIT的使用，請(qǐng)看TOM的解說(shuō)
http://asktom.oracle.com/pls/ask... 8_B:5918938803188,Y
do something like this: PHP code:
open cursor;
loop
fetch c bulk collect into l_c1, l_c2, ....... LIMIT 1000;
for i in 1 .. l_c1.count
loop
process....
end loop;
forall i in 1 .. l_c1.count
insert into ..... values ( L_c1(i), .... );
end loop;
exit when c%notfound;
end loop;
c
1000 would be extreme, 100 is reasonable.
聯(lián)系客服
