Question

这是从Tokutek下载的MariaDB + TokuDB 7.1社区。如果这是正常行为，请接受我的无知，但我对排序结果有疑问。我在两个排序方向之间进行排序时遇到了巨大的时间差异 - 升序和降序：

SELECT sql_no_cache id, createts, deleted
FROM sort_test
WHERE createts > '2000098'
ORDER BY createts asc

+---------+----------+---------+
| id      | createts | deleted |
+---------+----------+---------+
| 1999999 |  2000099 |    NULL |
| 2000000 |  2000100 |    NULL |
+---------+----------+---------+
2 rows in set (0.00 sec)

SELECT sql_no_cache id, createts, deleted
FROM sort_test
WHERE createts > '2000098'
ORDER BY createts desc

+---------+----------+---------+
| id      | createts | deleted |
+---------+----------+---------+
| 2000000 |  2000100 |    NULL |
| 1999999 |  2000099 |    NULL |
+---------+----------+---------+
2 rows in set (0.55 sec)

下面我介绍我的简化测试用例。这是表格：

CREATE TABLE `sort_test` (
  `id` int(11) NOT NULL AUTO_INCREMENT,
  `createts` int(11) DEFAULT NULL,
  `deleted` int(11) DEFAULT NULL,
  PRIMARY KEY (`id`),
  KEY `idx_createts` (`createts`)
) ENGINE=TokuDB

在这里，我使用以下过程在表中填充了200万行：

delimiter ;;

drop procedure if exists sort_test_populate;;

create procedure sort_test_populate()
begin
DECLARE int_val INT DEFAULT 1;
myloop : LOOP
  if (int_val > 2000000) THEN
    LEAVE myloop;
  end if;
  insert into sort_test (id, createts) values (int_val, int_val+100);
  set int_val = int_val +1;
end loop;
end;;

call sort_test_populate();;
Query OK, 1 row affected (28 min 2.80 sec)

以下是我的测试查询：

SELECT sql_no_cache id, createts, deleted
FROM sort_test
WHERE createts > '2000098'
ORDER BY createts asc

2 rows in set (0.00 sec)

SELECT sql_no_cache id, createts, deleted
FROM sort_test
WHERE createts > '2000098'
ORDER BY createts desc

2 rows in set (0.55 sec)

这是“解释扩展”的结果，它对于两个查询都是相同的：

+------+-------------+-----------+-------+---------------+--------------+---------+------+------+----------+-------------+
| id   | select_type | table     | type  | possible_keys | key          | key_len | ref  | rows | filtered | Extra       |
+------+-------------+-----------+-------+---------------+--------------+---------+------+------+----------+-------------+
|    1 | SIMPLE      | sort_test | range | idx_createts  | idx_createts | 5       | NULL |    2 |   100.00 | Using where |
+------+-------------+-----------+-------+---------------+--------------+---------+------+------+----------+-------------+

请注意，这不是我正在使用的确切数据，这里包含的内容太多了。我只是想创建一些测试数据来演示这个问题。我的问题是 - 为什么它的行为如此以及如何使降序查询更快？

Answer 1

这是索引条件下推（ICP）的已知错误。解决方法是通过全局或在执行此查询的会话中设置optimizer_switch来禁用ICP。

mysql> SET optimizer_switch='index_condition_pushdown=off';

（完全披露，我是Tokutek的员工，TokuDB的制造商）

TokuDB排序时间因ASC与DESC而异

1 个答案: