Why postgresql's single insert statement is faster than mysql? - mysql

Question

I wonder why postgresql's single "insert" statement is completely faster than MySQL's when autocommit is turned on? The following is the same code that I did on them.
Version:
MySQL: 5.6.10
PostgreSQL: PostgreSQL 9.3.2 on x86_64
Table definition:
MySQL:
CREATE TABLE `user` (
`username` char(36) NOT NULL,
`password` char(32) NOT NULL,
`register_time` datetime NOT NULL,
`mobile_phone` char(11) NOT NULL,
`is_admin` enum('yes','no') NOT NULL,
PRIMARY KEY (`username`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8
PostgreSQL:
CREATE TYPE ytt_enum AS ENUM ('yes','no');
CREATE TABLE ytt."user" (
"username" char(36) NOT NULL,
"password" char(32) NOT NULL,
"register_time" timestamp NOT NULL,
"mobile_phone" char(11) NOT NULL,
"is_admin" ytt_enum NOT NULL,
PRIMARY KEY ("username")
) ;
Store functions:
MySQL:
DELIMITER $$
USE `t_girl`$$
DROP PROCEDURE IF EXISTS `sp_insert_user_simple`$$
CREATE DEFINER=`root`#`localhost` PROCEDURE `sp_insert_user_simple`(
IN f_input INT
)
BEGIN
DECLARE i INT DEFAULT 0;
WHILE i <= f_input
DO
INSERT INTO t_girl.user (`username`, `password`, register_time,mobile_phone,is_admin)
VALUES (UUID(),MD5(REPLACE(UUID(),'-','')),DATE_SUB(NOW(),INTERVAL CEIL(RAND()*40) DAY),CEIL(RAND()*10000)[phone],IF(TRUNCATE(RAND()*2,0)=1,'yes','no'));
SET i = i + 1;
END WHILE;
END$$
DELIMITER ;
PostgreSQL:
CREATE or replace function sp_insert_user_simple(
IN f_input INT
) returns void as
$ytt$
declare i int := 0;
v_username char(36);
v_password char(32);
v_register_time timestamp;
v_mobile_phone char(11);
v_is_admin ytt_enum;
BEGIN
WHILE i < f_input
loop
v_username := uuid_generate_v1();
v_password :=MD5(REPLACE(uuid_generate_v1()::text,'-',''));
v_register_time := to_timestamp((now() - '1 day'::interval*ceil(random()*40))::text,'yyyy-mm-dd HH24:MI:SS');
v_mobile_phone :=CEIL(RANDOM()*10000)[phone];
v_is_admin := (case TRUNC(RANDOM()*2) when 1 then 'yes' else'no' end)::ytt_enum;
INSERT INTO ytt.user (username, password, register_time,mobile_phone,is_admin)
VALUES (v_username,v_password,v_register_time,v_mobile_phone,v_is_admin);
i := i + 1;
END loop;
END;
$ytt$language plpgsql;
Parameters:
MySQL:
innodb_buffer_pool_size=32M
bulk_insert_buffer_size=20M
autocommit=on
PostgreSQL:
shared_memory=32M
effective_cache_size=20M
autocommit=on
Test result:
MySQL:
mysql> call sp_insert_user_simple(10000);
Query OK, 1 row affected (1 min 9.93 sec)
PostgreSQL:
ytt=# select sp_insert_user_simple(10000);
sp_insert_user_simple
-----------------------
(1 row)
Time: 1177.043 ms
The above test shows that MySQL's running time is 69.93 second but PostgreSQL's is only 1.17 second.
Any answer is appreciated. Thanks.

Answer 1

I think what's happening here is that MySQL's procedures may be doing a commit for each individual INSERT. In PostgreSQL the whole procedure commits at the end; procedures cannot run individual transactions. (I'm not totally sure if that's how MySQL's procedures behave with autocommit=off, but it seems to be from a quick look at the docs).
You should really be doing this INSERT as a single statement anyway, using INSERT ... SELECT:
CREATE or replace function sp_insert_user_simple(
IN f_input integer
) returns void AS $$
INSERT INTO ytt.user (username, password, register_time,mobile_phone,is_admin)
SELECT
uuid_generate_v1(),
MD5(REPLACE(uuid_generate_v1()::text,'-','')),
to_timestamp((now() - '1 day'::interval*ceil(random()*40))::text,'yyyy-mm-dd HH24:MI:SS'),
CEIL(RANDOM()*10000)[phone],
case TRUNC(RANDOM()*2) when 1 then 'yes' else'no' end
FROM generate_series(1,$1);
$$ LANGUAGE sql;
(I assume this is dummy user-data generation?).
Also, use char, not varchar. char is an awful data type and should be avoided. Also, consider using boolean for the is_admin column.

Answer 2

Try testing simple insert queries:
INSERT INTO ytt.user (username, password) VALUES ('a', 'b');
and loop it in a procedure, thus making time measurement more accurate. Avoid using other built-in functions (like rng and timestamp), since their perfomance can differ singnificantly on a large sample, unless, of course, you tested those first.