我需要从一个大表(70M行)中获取最近的1000条记录,通过两个简单和小的表上的INNER JOIN匹配几个索引良好的项目。
查询需要1-2分钟才能运行。然而,explain
只能显示数百行。是什么给了什么?
如何优化查询或更有效地索引表,以便在我预期的毫秒内运行此查询?
表格:
score 70,000,000 records
class 400 records
category 400 records
查询:
SELECT
s.log_id,
s.category_id
FROM
score s
INNER JOIN category ca ON s.category_id = ca.id
INNER JOIN class cl ON ca.class_id = cl.id
WHERE
s.score_status_type_id = 0
AND ca.category_status_id = 1
AND cl.class_status_id IN (1, 2)
AND s.date > DATE_ADD(NOW(), INTERVAL -1440 minute)
GROUP BY s.log_id
ORDER BY s.date DESC
LIMIT 1000:
以下是解释:
*** row 1 ***
table: cl
type: range
possible_keys: PRIMARY,class_status_id
key: class_status_id
key_len: 4
ref: NULL
rows: 36
Extra: Using where; Using index; Using temporary; Using filesort
*** row 2 ***
table: ca
type: ref
possible_keys: PRIMARY,class_id,category_status_id,category_status_id_class_id_id
key: category_status_id_class_id_id
key_len: 8
ref: const,my_db.cl.id
rows: 1
Extra: Using index
*** row 3 ***
table: s
type: ref
possible_keys: unique_key,category_id,date,score,score_status_type_id,score_status_and_date,category_id_score_status_type_id_date_log_id,date_reverse,category_id_date_reverse,score_date
key: category_id_score_status_type_id_date_log_id
key_len: 8
ref: my_db.ca.id,const
rows: 396
Extra: Using where; Using index
以下是一些创建表:
CREATE TABLE `score` (
`log_id` bigint(20) NOT NULL,
`profile_id` bigint(20) DEFAULT NULL,
`date` datetime DEFAULT NULL,
`class_id` int(11) NOT NULL,
`score` float(10,6) DEFAULT NULL,
`score_date` datetime DEFAULT NULL,
`process_date` datetime DEFAULT NULL,
`status_type_id` int(3) NOT NULL DEFAULT '0',
`date_reverse` int(11) DEFAULT NULL,
UNIQUE KEY `unique_key` (`log_id`,`class_id`),
KEY `class_id` (`class_id`),
KEY `profile_id` (`profile_id`),
KEY `date` (`date`),
KEY `score` (`score`),
KEY `status_type_id` (`status_type_id `),
KEY `status_type_id_date` (`status_type_id`,`date`),
KEY `class_status_type_id_date_log_id` (`class_id`,`status_type_id`,`date`,`log_id`),
KEY `date_reverse` (`date_reverse`),
KEY `class_id_date_reverse` (`class_id`,`date_reverse`),
KEY `date` (`date`),
KEY `class_id_date_reverse_log_id` (`class_id`,`date_reverse`,`log_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
CREATE TABLE `category` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`class_id` int(11) NOT NULL,
`category_status_id` int(11) NOT NULL DEFAULT '0',
`name` varchar(255) COLLATE utf8_unicode_ci NOT NULL,
PRIMARY KEY (`id`),
KEY `class_id` (`class_id`),
KEY `name` (`name`),
KEY `category_status_id_class_id_id` (`category_status_id`,`class_id`,`id`)
) ENGINE=InnoDB AUTO_INCREMENT=412 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci
CREATE TABLE `class` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`class_status_id` int(11) NOT NULL DEFAULT '1',
`name` varchar(255) COLLATE utf8_unicode_ci NOT NULL,
PRIMARY KEY (`id`),
KEY `person_id` (`person_id`),
KEY `name` (`name`),
KEY `class_status_id` (`class_status_id`),
KEY `class_multi_1` (`class_status_id`,`name`,`id`)
) ENGINE=InnoDB AUTO_INCREMENT=407 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci
答案 0 :(得分:1)
问题是where
子句是过滤器,在连接完成后应用,因此你的连接表条件在哪里子句要求实际连接并将其放入临时结果集(可能很大)。通常,优化器会识别出条件可以在连接时断言,但有时它可能有点密集,所以......
尝试将非键条件移动到连接
SELECT s.log_id, s.category_id
FROM score s
JOIN category ca ON s.category_id = ca.id
AND ca.category_status_id = 1
JOIN class cl ON ca.class_id = cl.id
AND cl.class_status_id IN (1, 2)
WHERE s.score_status_type_id = 0
AND s.date > DATE_ADD(NOW(), INTERVAL -1440 minute)
GROUP BY s.log_id
ORDER BY s.date DESC
LIMIT 1000
如果这没有用,请尝试首先抓取score
行的子集作为子查询,然后进行联接。