我想运行一个简单的SELECT查询与基于PRIMARY KEY的WHERE语句。我在32核心和30 GB RAM的GCE实例上运行MySQL 5.7.9。 500个分区中有〜300M记录。我的查询中有99.9999%来自过去2天,而且这些数据完全符合内存,经过验证,我看到从磁盘读取约0个字节。我的CPU一直在100%挂起,其中10-20个同时运行的类似查询。MySQL的主键查询扫描5000x太多行
我正在跟踪一些Google搜索数据,因此每个组合的Requested
,IsPriority
和KeywordID
有100个rank
s。尽管我的主索引被设计为不需要扫描,但当我查看EXPLAIN
时,它显示正在扫描552k行以返回100行。
平均而言,从RAM中返回100个直接由主键指向的行需要20-40s。我能做些什么来加速这个查询?
EXPLAIN EXTENDED SELECT * FROM data.Rankings
-> WHERE Requested = '2016-02-26 00:00:00' AND NOT IsPriority AND KeywordID = '7387777811691965572'\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: Rankings
partitions: p20160226
type: ref
possible_keys: PRIMARY
key: PRIMARY
key_len: 4
ref: const
rows: 552598
filtered: 1.00
Extra: Using where
1 row in set, 2 warnings (0.01 sec)
这里的数据库表
CREATE TABLE `Rankings` (
`KeywordID` char(20) COLLATE utf8mb4_unicode_ci NOT NULL,
`Requested` timestamp NOT NULL,
`IsPriority` tinyint(1) NOT NULL,
`Retrieved` timestamp NOT NULL,
`Rank` tinyint(4) NOT NULL,
`Source` varchar(5) COLLATE utf8mb4_unicode_ci NOT NULL,
`ExternalID` varchar(45) COLLATE utf8mb4_unicode_ci NOT NULL,
`Phrase` varchar(255) COLLATE utf8mb4_unicode_ci NOT NULL,
// another 40 data columns of varying types
PRIMARY KEY (`Requested`,`IsPriority`,`KeywordID`,`Rank`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci ROW_FORMAT=COMPRESSED
// I have about 500 daily partitions over the last 1.5 years
/*!50100 PARTITION BY RANGE (UNIX_TIMESTAMP(Requested))
(PARTITION p20160222 VALUES LESS THAN (1456185600) ENGINE = InnoDB,
PARTITION p20160223 VALUES LESS THAN (1456272000) ENGINE = InnoDB,
PARTITION p20160224 VALUES LESS THAN (1456358400) ENGINE = InnoDB,
PARTITION p20160225 VALUES LESS THAN (1456444800) ENGINE = InnoDB,
PARTITION p20160226 VALUES LESS THAN (1456531200) ENGINE = InnoDB,
PARTITION p20160227 VALUES LESS THAN (1456617600) ENGINE = InnoDB,
PARTITION p20160228 VALUES LESS THAN (1456704000) ENGINE = InnoDB,
PARTITION p20160229 VALUES LESS THAN (1456790400) ENGINE = InnoDB) */;
我得仔细看才能提供一个可靠的答案。我通常在索引中避免复合键 - 查询优化器必须做很多工作才能确定它是否可以将PK用作索引。相反,请考虑在每个字段上创建多个索引。 –