使用索引的慢 MySQL 查询

Question

我有一个 Providers 表，如下所示：

|编号 |姓氏 |名字 |中间名 | | --- | -------- | --------- | ---------- |

具有以下索引：

• Providers_lastName
• Providers_firstName
• Providers_lastName_firstName
• Providers_lastName_firstName_middleName

我的所有查询在 lastName 和 firstName 值中都使用尾随通配符：

SELECT * FROM Providers
WHERE lastName LIKE 'smi%'
ORDER BY lastName ASC, firstName ASC, middleName
LIMIT 0, 50

SELECT * FROM Providers
WHERE firstName LIKE 'mar%'
ORDER BY lastName ASC, firstName ASC, middleName
LIMIT 0, 50

我在这个表中有大约 700 万行。我的姓氏查询非常快。但是，firstName 的速度非常慢。我在这里做错了什么吗？在不更改或删除顺序的情况下，我还可以添加哪些其他索引来提高仅 firstName 查询的性能？

编辑 1：

EXPLAINlastName 查询的输出：

{
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "69901.30"
    },
    "ordering_operation": {
      "using_filesort": false,
      "table": {
        "table_name": "Providers",
        "access_type": "range",
        "possible_keys": [
          "Providers_lastName",
          "Providers_lastName_firstName",
          "Providers_lastName_firstName_middleName"
        ],
        "key": "Providers_lastName_firstName_middleName",
        "used_key_parts": [
          "lastName"
        ],
        "key_length": "143",
        "rows_examined_per_scan": 59008,
        "rows_produced_per_join": 59008,
        "filtered": "100.00",
        "index_condition": "(`db_name`.`providers`.`lastName` like 'smi%')",
        "cost_info": {
          "read_cost": "64000.51",
          "eval_cost": "5900.80",
          "prefix_cost": "69901.31",
          "data_read_per_join": "158M"
        },
        "used_columns": [
          "id",
          "firstName",
          "middleName",
          "lastName",
          // OTHER COLUMNS
        ]
      }
    }
  }
}

EXPLAINfirstName 查询的输出：

{
  "query_block": {
    "select_id": 1,
    "cost_info": {
      "query_cost": "390813.95"
    },
    "ordering_operation": {
      "using_filesort": false,
      "table": {
        "table_name": "Providers",
        "access_type": "index",
        "possible_keys": [
          "Providers_firstName"
        ],
        "key": "Providers_lastName_firstName_middleName",
        "used_key_parts": [
          "lastName",
          "firstName",
          "middleName"
        ],
        "key_length": "309",
        "rows_examined_per_scan": 948,
        "rows_produced_per_join": 329914,
        "filtered": "5.27",
        "cost_info": {
          "read_cost": "357822.55",
          "eval_cost": "32991.40",
          "prefix_cost": "390813.95",
          "data_read_per_join": "883M"
        },
        "used_columns": [
          "id",
          "firstName",
          "middleName",
          "lastName",
          // OTHER COLUMNS
        ],
        "attached_condition": "(`db_name`.`providers`.`firstName` like 'mar%')"
      }
    }
  }
}

SHOW CREATE TABLE:

CREATE TABLE `Providers` (
  `id` varchar(10) NOT NULL,
  `firstName` varchar(20) DEFAULT NULL,
  `middleName` varchar(20) DEFAULT NULL,
  `lastName` varchar(35) DEFAULT NULL,
  /* Other columns */
  PRIMARY KEY (`id`),
  KEY `Providers_firstName` (`firstName`),
  KEY `Providers_lastName` (`lastName`),
  KEY `Providers_lastName_firstName` (`lastName`,`firstName`),
  KEY `Providers_lastName_firstName_middleName` (`lastName`,`firstName`,`middleName`),
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci

编辑 2：

运行FLUSH STATUS 后SHOW SESSION STATUS LIKE 'Handler%' 的输出：

查询 1（名字）：

{
    "data":
    [
        {
            "Variable_name": "Handler_commit",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_delete",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_discover",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_external_lock",
            "Value": "2"
        },
        {
            "Variable_name": "Handler_mrr_init",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_prepare",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_first",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_read_key",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_read_last",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_next",
            "Value": "1487176"
        },
        {
            "Variable_name": "Handler_read_prev",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_rnd",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_rnd_next",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_update",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_write",
            "Value": "0"
        }
    ]
}

查询 2（姓氏）：

{
    "data":
    [
        {
            "Variable_name": "Handler_commit",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_delete",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_discover",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_external_lock",
            "Value": "2"
        },
        {
            "Variable_name": "Handler_mrr_init",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_prepare",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_first",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_key",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_read_last",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_next",
            "Value": "49"
        },
        {
            "Variable_name": "Handler_read_prev",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_rnd",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_rnd_next",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_update",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_write",
            "Value": "0"
        }
    ]
}

编辑 3

使用FORCE_INDEX(Providers_firstName):

EXPLAINfirstName 查询的输出：

{
    "query_block": {
      "select_id": 1,
      "cost_info": {
        "query_cost": "389514.60"
      },
    "ordering_operation": {
        "using_filesort": true,
        "table": {
          "table_name": "Providers",
          "access_type": "range",
          "possible_keys": [
            "Providers_firstName"
          ],
          "key": "Providers_firstName",
          "used_key_parts": [
            "firstName"
          ],
          "key_length": "83",
          "rows_examined_per_scan": 329914,
          "rows_produced_per_join": 329914,
          "filtered": "100.00",
          "index_condition": "(`db_name`.`providers`.`firstName` like 'mar%')",
          "cost_info": {
            "read_cost": "356523.20",
            "eval_cost": "32991.40",
            "prefix_cost": "389514.60",
            "data_read_per_join": "883M"
          },
        "used_columns": [
            "id",
            "firstName",
            "middleName",
            "lastName",
            // Other columns
          ]
      }
    }
  }
}

处理程序计数：

{
    "data":
    [
        {
            "Variable_name": "Handler_commit",
            "Value": "1"
        },
        {
            "Variable_name": "Handler_delete",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_discover",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_external_lock",
            "Value": "2"
        },
        {
            "Variable_name": "Handler_mrr_init",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_prepare",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_first",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_key",
            "Value": "51"
        },
        {
            "Variable_name": "Handler_read_last",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_next",
            "Value": "168497"
        },
        {
            "Variable_name": "Handler_read_prev",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_read_rnd",
            "Value": "50"
        },
        {
            "Variable_name": "Handler_read_rnd_next",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_savepoint_rollback",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_update",
            "Value": "0"
        },
        {
            "Variable_name": "Handler_write",
            "Value": "0"
        }
    ]
}

Answer 1

查询 1

WHERE lastName LIKE 'smi%'
ORDER BY lastName ASC, firstName ASC, middleName

可能使用此索引。 （请提供EXPLAIN...）：

Providers_lastName_firstName_middleName

这样比较有效，因为它可以遍历索引的smi... 部分。

我假设 SELECT * 只获取 4 列，而 id 是 PRIMARY KEY?? Providers_lastName_firstName_middleName 是 INDEX(lastName, firstName, middleName)，最后加上一个隐含的 id，因为它是 InnoDB？？

这意味着整个查询可以在索引中运行。 EXPLAIN 会通过说“使用索引”来确认这一点，这意味着“覆盖索引”。

此外，此查询仅涉及 50 行 - 因为索引非常适合 WHERE 和 ORDER BY，它实际上也可以折叠在 LIMIT 50 中。

查询 2

WHERE firstName LIKE 'mar%'
ORDER BY lastName ASC, firstName ASC, middleName

Providers_firstName

也可以遍历mar... 的索引，但随后必须访问数据以获取其余列。

但其他优化（覆盖等）均不适用。您可以添加INDEX(first, last, middle, id) 以使其更快。

此查询无法折叠到LIMIT。

备注

在美国，10% 的名称以最常见的字母“S”开头。 （“10%”在全球范围内大致相同，但最受欢迎的字母可能会有所不同。）

优化器有多种方法来执行任何查询，并根据有限的信息选择“最佳”。当一个范围很明显时 (WHERE lastName LIKE 'S%')，它可能选择从使用索引切换到简单地丢弃许多行。我认为这不会发生在这里，但EXPLAIN 会再次告诉我们。

有关创建最佳索引的更多信息：http://mysql.rjweb.org/doc.php/index_cookbook_mysql

解释后

如果我正确阅读了EXPLAINs，他们都使用INDEX(last, first, middle)，从而避免了排序。另请注意"using_filesort": false.，这允许查询在LIMIT 50 之后停止。

要收集更多信息，请运行以下命令：

FLUSH STATUS;
SELECT ...
SHOW SESSION STATUS LIKE 'Handler%';

如果Handler_write* 是0，则没有排序。同时，被触摸的Handler_read* values gives you the number of rows (probably in theINDEX`) 的总和。

我希望查询 1 总共显示 50 个读取，因为它（理论上）可以深入到 smi 处的索引并获取接下来的 50 个（或更少）行。这应该只需要几毫秒。

查询 2 比较混乱，因为它需要扫描大量索引才能找到具有该名字的 50 个索引。它不会是 7M，但可能是 50K 行。如果索引的必要部分被缓存，这可能需要几秒钟；如果不是，则等待几分钟。

没有办法让第二季度的速度和第一季度一样快。这可能对于mar% 更快，但对于m% 更慢：INDEX(first, last, middle)。也就是说，引入这样的索引是有风险的。

在大多数情况下，如果您还拥有INDEX(a,b)，INDEX(a) 是多余的。也就是说，您有 2 个可能会被删除的索引。