MySQL 性能较慢 *with* 索引？

Question

我正在编写一个应用程序，该应用程序使用 MySQL 将文件哈希数据保存到具有单个表的简单数据库中。我创建如下:

CREATE DATABASE IF NOT EXISTS hash_db;

CREATE TABLE IF NOT EXISTS hash_db.main_tbl  
(                                       
  sha256       CHAR(64) PRIMARY KEY    ,
  sha1         CHAR(40) UNIQUE KEY     ,
  md5          CHAR(32) UNIQUE KEY     ,
  created      DATETIME                ,
  modified     DATETIME                ,
  size         BIGINT                  ,
  ext          VARCHAR(260)            ,
  path         TEXT(32768)             ,
  new_record   BOOL                     
 )                                      
ENGINE = MyISAM

CREATE UNIQUE INDEX sha256_idx ON hash_db.main_tbl (sha256)
CREATE UNIQUE INDEX sha1_idx   ON hash_db.main_tbl (sha1)
CREATE UNIQUE INDEX md5_idx    ON hash_db.main_tbl (md5)

然后我只 进行简单的表单选择和插入:

SELECT * FROM hash_db.main_tbl WHERE
      sha256 = '...'   OR
      sha1   = '...'   OR
      md5    = '...'


INSERT INTO hash_db.main_tbl
  (sha256, sha1, md5, created, modified, size, ext, path, new_record) VALUES
  (
    '...'                    ,
    '...'                    ,
    '...'                    ,
    FROM_UNIXTIME(...)       ,
    FROM_UNIXTIME(...)       ,
    ...                      ,
    '...'                    ,
    '...'                    ,
    TRUE                                  
  )

数据几乎是随机的，唯一性概率非常高(不是说它应该重要，还是应该重要？)。 第一个问题，对于这种用法，InnoDB 比 MyISAM 慢得多(慢 7 倍)是否正常？我读到它应该是相反的(尝试使用 512M innodb_buffer_pool_size，没有区别)。

其次...我已经测试了有无索引(MyISAM)，有索引的版本实际上更慢。这些是我的应用测量的实际性能数据(使用 C 中的性能计数器):

With indexes:
Selects per second: 393.7
Inserts per second: 1056.1

Without indexes:
Selects per second: 585.3
Inserts per second: 1480.9

我得到的数据是可重复的。我已经测试了扩大的 key_buffer_size(32M，默认为 8M)。

我做错了什么或遗漏了什么？

============================================= =================================

根据 Gordon Linoff 的建议编辑:

我尝试过使用 UNION ALL，但实际上性能下降了，每秒精确选择 70 次。 EXPLAIN 的输出如下:

EXPLAIN EXTENDED SELECT * FROM main_hash_db.main_tbl WHERE md5 = '...'

+----+-------------+----------+-------+---------------+------+---------+-------+------+----------+-------+
| id | select_type | table    | type  | possible_keys | key  | key_len | ref   | rows | filtered | Extra |
+----+-------------+----------+-------+---------------+------+---------+-------+------+----------+-------+
|  1 | SIMPLE      | main_tbl | const | md5           | md5  | 97      | const |    1 |   100.00 | NULL  |
+----+-------------+----------+-------+---------------+------+---------+-------+------+----------+-------+


EXPLAIN EXTENDED SELECT * FROM main_hash_db.main_tbl WHERE md5 = '...' UNION ALL SELECT * FROM main_hash_db.main_tbl WHERE sha1 = '...'

+----+--------------+------------+-------+-----------------------+------+---------+-------+------+----------+-----------------+
| id | select_type  | table      | type  | possible_keys         | key  | key_len | ref   | rows | filtered | Extra           |
+----+--------------+------------+-------+-----------------------+------+---------+-------+------+----------+-----------------+
|  1 | PRIMARY      | main_tbl   | const | md5                   | md5  | 97      | const |    1 |   100.00 | NULL            |
|  2 | UNION        | main_tbl   | const | sha1,sha1_idx,md5_idx | sha1 | 121     | const |    1 |   100.00 | NULL            |
| NULL | UNION RESULT | <union1,2> | ALL   | NULL                  | NULL | NULL    | NULL  | NULL |     NULL | Using temporary |
+----+--------------+------------+-------+-----------------------+------+---------+-------+------+----------+-----------------+


EXPLAIN EXTENDED SELECT * FROM main_hash_db.main_tbl WHERE md5 = '...' UNION ALL SELECT * FROM main_hash_db.main_tbl WHERE sha1 = '...' UNION ALL SELECT * FROM main_hash_db.main_tbl WHERE sha256 = '...'

+----+--------------+--------------+-------+-----------------------+---------+---------+-------+------+----------+-----------------+
| id | select_type  | table        | type  | possible_keys         | key     | key_len | ref   | rows | filtered | Extra           |
+----+--------------+--------------+-------+-----------------------+---------+---------+-------+------+----------+-----------------+
|  1 | PRIMARY      | main_tbl     | const | md5                   | md5     | 97      | const |    1 |   100.00 | NULL            |
|  2 | UNION        | main_tbl     | const | sha1,sha1_idx,md5_idx | sha1    | 121     | const |    1 |   100.00 | NULL            |
|  3 | UNION        | main_tbl     | const | PRIMARY,sha256_idx    | PRIMARY | 192     | const |    1 |   100.00 | NULL            |
| NULL | UNION RESULT | <union1,2,3> | ALL   | NULL                  | NULL    | NULL    | NULL  | NULL |     NULL | Using temporary |
+----+--------------+--------------+-------+-----------------------+---------+---------+-------+------+----------+-----------------+

这让我发现我在创建索引时出错(我正在为“sha1”列创建两个单独的索引)。但在修复之后，速度仍然很慢(每秒约 70 次选择)，这是 EXPLAIN 的输出:

+----+--------------+--------------+-------+--------------------+---------+---------+-------+------+----------+-----------------+
| id | select_type  | table        | type  | possible_keys      | key     | key_len | ref   | rows | filtered | Extra           |
+----+--------------+--------------+-------+--------------------+---------+---------+-------+------+----------+-----------------+
|  1 | PRIMARY      | main_tbl     | const | md5,md5_idx        | md5     | 97      | const |    1 |   100.00 | NULL            |
|  2 | UNION        | main_tbl     | const | sha1,sha1_idx      | sha1    | 121     | const |    1 |   100.00 | NULL            |
|  3 | UNION        | main_tbl     | const | PRIMARY,sha256_idx | PRIMARY | 192     | const |    1 |   100.00 | NULL            |
| NULL | UNION RESULT | <union1,2,3> | ALL   | NULL               | NULL    | NULL    | NULL  | NULL |     NULL | Using temporary |
+----+--------------+--------------+-------+--------------------+---------+---------+-------+------+----------+-----------------+

============================================= =================================

经过进一步讨论后的第三次编辑(见下文)。这是原始查询的 EXPLAIN 输出(没有定义额外的索引，数据库是如上所述创建的):

explain extended select path from main_hash_db.main_tbl where sha256 = '...' or md5 = '...' or sha1 = '...' ;

+----+-------------+----------+-------------+------------------+------------------+------------+------+------+----------+--------------------------------------------+
| id | select_type | table    | type        | possible_keys    | key              | key_len    | ref  | rows | filtered | Extra                                      |
+----+-------------+----------+-------------+------------------+------------------+------------+------+------+----------+--------------------------------------------+
|  1 | SIMPLE      | main_tbl | index_merge | PRIMARY,sha1,md5 | PRIMARY,md5,sha1 | 192,97,121 | NULL |    3 |   100.00 | Using union(PRIMARY,md5,sha1); Using where |
+----+-------------+----------+-------------+------------------+------------------+------------+------+------+----------+--------------------------------------------+

我的应用衡量的性能:

Selects per second: 500.6
Inserts per second: 1394.8

这是 3 次选择的结果(单独发布，而不是 UNION):

Selects per second: 2525.1
Inserts per second: 1584.3

Answer 1

首先，您会期望没有索引的 insert 会更快。那里没有什么神秘之处。不必维护该索引。事实上，在进行大插入时，一个好的策略通常是先删除索引，进行插入，然后重建索引。

select比较麻烦。毕竟，这是您希望使用索引的地方。您的查询是:

SELECT *
FROM hash_db.main_tbl
WHERE sha256 = '...'   OR
      sha1   = '...'   OR
      md5    = '...';

这恰好是索引使用的最坏情况。您需要查看 explain 以了解如何使用索引。

我的建议是这样写查询:

SELECT *
FROM hash_db.main_tbl
WHERE sha256 = '...'
UNION ALL
SELECT *
FROM hash_db.main_tbl
sha1   = '...'
UNION ALL
SELECT *
FROM hash_db.main_tbl
WHERE md5    = '...';

(或者如果你真的想消除重复项，请使用 union。)

这应该利用每个子查询的每个索引，并且应该为您提供所需的性能。