当涉及范围时,索引中的第一个更高的基数列?

时间:2018-05-08 17:59:23

标签: mysql performance indexing query-optimization mariadb

CREATE TABLE `files` (
  `did` int(10) unsigned NOT NULL DEFAULT '0',
  `filename` varbinary(200) NOT NULL,
  `ext` varbinary(5) DEFAULT NULL,
  `fsize` double DEFAULT NULL,
  `filetime` datetime DEFAULT NULL,
  PRIMARY KEY (`did`,`filename`),
  KEY `fe` (`filetime`,`ext`),          -- This?
  KEY `ef` (`ext`,`filetime`)           -- or This?
) ENGINE=InnoDB DEFAULT CHARSET=utf8 ;

表格中有一百万行。文件时间大多不同。有一定数量的ext值。因此,filetime具有较高的基数,ext的基数较低。

该查询涉及extfiletime

WHERE ext = '...'
  AND filetime BETWEEN ... AND ...

这两个指标中的哪一个更好?为什么?

1 个答案:

答案 0 :(得分:6)

首先,让我们FORCE INDEX选择effe。时间太短,无法清楚地了解哪个更快,但`EXPLAIN显示出差异:

首先强制filetime上的范围。 (注意:WHERE中的顺序没有影响。)

mysql> EXPLAIN SELECT COUNT(*), AVG(fsize)
    FROM files FORCE INDEX(fe)
    WHERE ext = 'gif' AND filetime >= '2015-01-01'
                      AND filetime <  '2015-01-01' + INTERVAL 1 MONTH;
+----+-------------+-------+-------+---------------+------+---------+------+-------+-----------------------+
| id | select_type | table | type  | possible_keys | key  | key_len | ref  | rows  | Extra                 |
+----+-------------+-------+-------+---------------+------+---------+------+-------+-----------------------+
|  1 | SIMPLE      | files | range | fe            | fe   | 14      | NULL | 16684 | Using index condition |
+----+-------------+-------+-------+---------------+------+---------+------+-------+-----------------------+

首先强制低基数ext

mysql> EXPLAIN SELECT COUNT(*), AVG(fsize)
    FROM files FORCE INDEX(ef)
    WHERE ext = 'gif' AND filetime >= '2015-01-01'
                      AND filetime <  '2015-01-01' + INTERVAL 1 MONTH;
+----+-------------+-------+-------+---------------+------+---------+------+------+-----------------------+
| id | select_type | table | type  | possible_keys | key  | key_len | ref  | rows | Extra                 |
+----+-------------+-------+-------+---------------+------+---------+------+------+-----------------------+
|  1 | SIMPLE      | files | range | ef            | ef   | 14      | NULL |  538 | Using index condition |
+----+-------------+-------+-------+---------------+------+---------+------+------+-----------------------+

显然,rowsef更好。但是,让我们检查优化器跟踪。产量相当笨重;我只展示有趣的部分。不需要FORCE;跟踪将显示两个选项然后选择更好。

             ...
             "potential_range_indices": [
                ...
                {
                  "index": "fe",
                  "usable": true,
                  "key_parts": [
                    "filetime",
                    "ext",
                    "did",
                    "filename"
                  ]
                },
                {
                  "index": "ef",
                  "usable": true,
                  "key_parts": [
                    "ext",
                    "filetime",
                    "did",
                    "filename"
                  ]
                }
              ],

...

              "analyzing_range_alternatives": {
                "range_scan_alternatives": [
                  {
                    "index": "fe",
                    "ranges": [
                      "2015-01-01 00:00:00 <= filetime < 2015-02-01 00:00:00"
                    ],
                    "index_dives_for_eq_ranges": true,
                    "rowid_ordered": false,
                    "using_mrr": false,
                    "index_only": false,
                    "rows": 16684,
                    "cost": 20022,               <-- Here's the critical number
                    "chosen": true
                  },
                  {
                    "index": "ef",
                    "ranges": [
                      "gif <= ext <= gif AND 2015-01-01 00:00:00 <= filetime < 2015-02-01 00:00:00"
                    ],
                    "index_dives_for_eq_ranges": true,
                    "rowid_ordered": false,
                    "using_mrr": false,
                    "index_only": false,
                    "rows": 538,
                    "cost": 646.61,               <-- Here's the critical number
                    "chosen": true
                  }
                ],

...

          "attached_conditions_computation": [
            {
              "access_type_changed": {
                "table": "`files`",
                "index": "ef",
                "old_type": "ref",
                "new_type": "range",
                "cause": "uses_more_keyparts"   <-- Also interesting
              }
            }

使用fe(范围列第一个),可以使用范围,但估计扫描16684行捕获ext='gif'

使用ef(低基数ext优先),它可以使用索引的两列并在BTree中更有效地向下钻取。然后它发现了大约538行,所有这些行对查询都很有用 - 不需要进一步过滤。

结论:

  • INDEX(filetime, ext)仅使用第一列。
  • INDEX(ext, filetime)使用了两列。
  • =测试首先中涉及的列放在索引中,而不考虑基数
  • 查询计划不会超出第一个范围&#39;列。
  • &#34;基数&#34;与复合索引和此类查询无关

(&#34;使用索引条件&#34;意味着存储引擎(InnoDB)将使用超出用于过滤的索引的列。&#34;)