[英]Query with MySQL join is slow, is there a way to improve slow performance with subquery?
我正在使用 InnoDB 的全文搜索从 2 个表格、照片和相册中检索结果,如下所示:
SELECT p.media_extension,
p.media_vishash,
ANY_VALUE(a.title_url) as album_title_url,
ANY_VALUE(p.title_url) as photo_title_url,
p.media_time_created,
p.comments,
p.title,
p.text,
MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename)
AGAINST ('search query' IN BOOLEAN MODE) as score
FROM albums a,
album_photo_map apm,
media p
WHERE p.media_vishash = apm.media_vishash AND
apm.album_id = a.album_id AND
p.level <= 5 AND
a.level <= 5 AND
MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename)
AGAINST ('search query' IN BOOLEAN MODE)
GROUP BY p.media_vishash
ORDER BY p.datetime_created DESC, p.media_vishash DESC
在 0.80 秒内执行。
+----+-------------+-------+------------+----------+-----------------------------------------+----------+---------+--------------------+------+----------+----------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+----------+-----------------------------------------+----------+---------+--------------------+------+----------+----------------------------------------------+
| 1 | SIMPLE | a | NULL | ALL | PRIMARY,album_id_type,album_level,level | NULL | NULL | NULL | 9147 | 49.99 | Using where; Using temporary; Using filesort |
| 1 | SIMPLE | p | NULL | fulltext | media_vishash,search | search | 0 | const | 1 | 33.33 | Using where |
| 1 | SIMPLE | apm | NULL | ref | PRIMARY,album_id,media_vishash | album_id | 130 | lifebox.a.album_id | 26 | 10.00 | Using where; Using index |
+----+-------------+-------+------------+----------+-----------------------------------------+----------+---------+--------------------+------+----------+----------------------------------------------+
与专辑表的连接确实减慢了查询速度。 我尝试在没有加入的情况下重写它,如下所示,性能非常好。
SELECT p.media_extension,
p.media_vishash,
ANY_VALUE(p.title_url) as photo_title_url,
p.media_time_created,
p.comments,
p.title,
p.text,
MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename) AGAINST ('search query' IN BOOLEAN MODE) as score
FROM media p
WHERE MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename)
AGAINST ('search query' IN BOOLEAN MODE)
在 0.01 秒内执行。
问题是我需要从专辑表中检索 title_url,而第二个查询中缺少此内容。
有没有一种方法可以像在第一个查询中一样检索 Albums.title_url,同时保持第二个查询的出色性能,而无需将其拆分为多个查询? 也许通过子选择从专辑表中获取 title_url?
创建下表,MySQL 8.0.27:
CREATE TABLE `media` (
`media_id` char(22) CHARACTER SET utf8mb4 COLLATE utf8mb4_bin NOT NULL,
`media_sha512` char(128) NOT NULL,
`media_sha256` char(64) NOT NULL,
`media_md5` char(32) NOT NULL,
`media_filepath` varchar(250) NOT NULL,
`media_vishash` char(128) NOT NULL,
`media_filename` varchar(250) NOT NULL,
`media_extension` varchar(10) NOT NULL,
`media_path` varchar(250) NOT NULL,
`media_folder` varchar(250) NOT NULL,
`media_time_lastmod` bigint NOT NULL,
`media_time_created` bigint DEFAULT NULL,
`media_filesize` bigint DEFAULT NULL,
`media_width` int DEFAULT NULL,
`media_height` int DEFAULT NULL,
`media_megapixels` float DEFAULT NULL,
`media_mimetype` varchar(50) DEFAULT NULL,
`media_type` varchar(25) DEFAULT NULL,
`title` varchar(250) NOT NULL,
`title_url` varchar(250) NOT NULL,
`title_url_previous` varchar(250) DEFAULT NULL,
`text` text,
`year` int DEFAULT NULL,
`day` int DEFAULT NULL,
`month` int DEFAULT NULL,
`sidecarimg_filepath` varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`sidecarimg_md5` char(32) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`sidecarvideo_filepath` varchar(250) DEFAULT NULL,
`sidecarvideo_md5` char(32) DEFAULT NULL,
`sidecarxmp_filepath` varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`sidecarxmp_md5` char(32) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`modified_filepath` varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`modified_md5` char(32) DEFAULT NULL,
`modified_time_lastmod` int DEFAULT NULL,
`modified_time_created` int DEFAULT NULL,
`datetime_created` datetime DEFAULT NULL,
`datetime_modified` datetime DEFAULT NULL,
`EXIF_Make` varchar(50) DEFAULT NULL,
`EXIF_LensModel` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`EXIF_LensID` varchar(100) DEFAULT NULL,
`EXIF_LensMount` varchar(100) DEFAULT NULL,
`EXIF_LensFormat` varchar(100) DEFAULT NULL,
`EXIF_Software` varchar(100) DEFAULT NULL,
`EXIF_ByLine` varchar(100) DEFAULT NULL,
`EXIF_Copyright` varchar(100) DEFAULT NULL,
`EXIF_DateTimeOriginal` datetime DEFAULT NULL,
`EXIF_ExposureTime` varchar(100) DEFAULT NULL,
`EXIF_ShutterSpeed` varchar(100) DEFAULT NULL,
`EXIF_FNumber` varchar(100) DEFAULT NULL,
`EXIF_FocalLength` varchar(100) DEFAULT NULL,
`EXIF_FocalLength35mm` varchar(100) DEFAULT NULL,
`EXIF_ISO` varchar(100) DEFAULT NULL,
`EXIF_DisplayProfile` varchar(100) DEFAULT NULL,
`EXIF_Keywords` varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`EXIF_Model` varchar(75) DEFAULT NULL,
`EXIF_Orientation` varchar(100) DEFAULT NULL,
`EXIF_ExposureProgram` varchar(100) DEFAULT NULL,
`EXIF_BitsPerSample` varchar(100) DEFAULT NULL,
`EXIF_FlashFire` varchar(100) DEFAULT NULL,
`EXIF_ExposureCompensation` varchar(100) DEFAULT NULL,
`EXIF_HDR` varchar(100) DEFAULT NULL,
`EXIF_ColorTemperature` varchar(25) DEFAULT NULL,
`EXIF_ColorSpace` varchar(20) DEFAULT NULL,
`EXIF_ColorProfileDescription` varchar(50) DEFAULT NULL,
`EXIF_CustomRendered` varchar(100) DEFAULT NULL,
`EXIF_HistorySoftwareAgent` varchar(500) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`EXIF_WhiteBalance` varchar(100) DEFAULT NULL,
`EXIF_Quality` varchar(20) DEFAULT NULL,
`EXIF_SelfTimer` varchar(100) DEFAULT NULL,
`EXIF_Contrast` varchar(100) DEFAULT NULL,
`EXIF_Saturation` varchar(100) DEFAULT NULL,
`EXIF_Sharpness` varchar(100) DEFAULT NULL,
`EXIF_SerialNumber` varchar(100) DEFAULT NULL,
`EXIF_ShutterCount` varchar(100) DEFAULT NULL,
`EXIF_Rating` varchar(100) DEFAULT NULL,
`EXIF_RatingPercent` varchar(100) DEFAULT NULL,
`EXIF_Good` tinyint DEFAULT NULL,
`EXIF_Subject` varchar(250) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`EXIF_Title` varchar(250) DEFAULT NULL,
`EXIF_Description` text CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci,
`EXIF_Location` varchar(250) DEFAULT NULL,
`GPS_Altitude_Ref` varchar(50) DEFAULT NULL,
`GPS_Altitude` varchar(50) DEFAULT NULL,
`GPS_Latitude` varchar(50) DEFAULT NULL,
`GPS_Longitude` varchar(50) DEFAULT NULL,
`GPS_Latitude_Dec` double DEFAULT NULL,
`GPS_Longitude_Dec` double DEFAULT NULL,
`GPS_datetime` datetime DEFAULT NULL,
`is_video` tinyint(1) NOT NULL DEFAULT '0',
`video_duration` varchar(50) DEFAULT NULL,
`video_framerate` varchar(50) DEFAULT NULL,
`video_bitrate` varchar(50) DEFAULT NULL,
`video_resolution` varchar(5) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT NULL,
`video_codec` varchar(50) DEFAULT NULL,
`video_pixfmt` varchar(50) DEFAULT NULL,
`video_profile` varchar(50) DEFAULT NULL,
`video_colorprimary` varchar(50) DEFAULT NULL,
`video_rotation` varchar(5) DEFAULT NULL,
`audio_codec` varchar(50) DEFAULT NULL,
`audio_bitrate` varchar(50) DEFAULT NULL,
`audio_channels` varchar(50) DEFAULT NULL,
`audio_bits_per_sample` varchar(50) DEFAULT NULL,
`audio_sample_rate` varchar(50) DEFAULT NULL,
`comment_list` text,
`tag_list` text,
`timestamp_last_comment` int NOT NULL DEFAULT '0',
`timestamp_last_view` int DEFAULT NULL,
`views` int NOT NULL DEFAULT '0',
`comments` int NOT NULL DEFAULT '0',
`tags` int NOT NULL DEFAULT '0',
`thumbs` json DEFAULT NULL,
`thumbs_datetime_created` datetime DEFAULT NULL,
`level` tinyint NOT NULL DEFAULT '2',
PRIMARY KEY (`media_id`),
UNIQUE KEY `media_vishash` (`media_vishash`) USING BTREE,
UNIQUE KEY `media_sha512` (`media_sha512`) USING BTREE,
KEY `index_rating` (`EXIF_Rating`),
FULLTEXT KEY `search` (`title`,`text`,`tag_list`,`comment_list`,`media_filename`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci
CREATE TABLE `album_photo_map` (
`album_id` varchar(32) NOT NULL DEFAULT '',
`media_vishash` varchar(128) CHARACTER SET utf8 COLLATE utf8_bin NOT NULL,
PRIMARY KEY (`album_id`,`media_vishash`) USING BTREE,
KEY `album_id` (`album_id`),
KEY `media_vishash` (`media_vishash`) USING BTREE
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci
CREATE TABLE `albums` (
`album_id` varchar(32) NOT NULL DEFAULT '',
`title` varchar(100) NOT NULL DEFAULT '',
`title_url` varchar(100) NOT NULL DEFAULT '',
`location` varchar(100) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT '',
`text` text,
`timestamp_created` int NOT NULL DEFAULT '0',
`timestamp_modified` int DEFAULT '0',
`timestamp_oldest` int DEFAULT NULL,
`timestamp_newest` int DEFAULT NULL,
`num_photos` int NOT NULL DEFAULT '0',
`num_videos` int NOT NULL DEFAULT '0',
`num_items` int NOT NULL DEFAULT '0',
`type` tinyint(1) NOT NULL DEFAULT '1',
`highlight_media_vishash` varchar(128) CHARACTER SET utf8 COLLATE utf8_bin DEFAULT NULL,
`timestamp_last_view` int NOT NULL DEFAULT '0',
`last_viewer` varchar(32) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci DEFAULT '',
`views` int NOT NULL DEFAULT '0',
`min_level` tinyint(1) NOT NULL DEFAULT '2',
`level` tinyint(1) NOT NULL DEFAULT '2',
PRIMARY KEY (`album_id`),
UNIQUE KEY `title` (`title`),
UNIQUE KEY `title_url` (`title_url`),
UNIQUE KEY `album_id_type` (`album_id`,`type`),
KEY `highlight_photo_id` (`highlight_media_vishash`),
KEY `lowest_level` (`min_level`),
KEY `timestamp_created` (`timestamp_created`),
KEY `type` (`type`),
KEY `level` (`level`),
FULLTEXT KEY `title_2` (`title`,`text`,`location`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci
我使用 EXPLAIN 测试了您的第一个查询,发现优化器想要重新排序表,因此它以album_photo_map
,然后连接到其他每个表的主键。 这意味着它根本没有使用您的全文索引。
因此,我使用STRAIGHT_JOIN
语法测试了查询,以覆盖优化器对表重新排序的趋势。 这也导致我使用现代的JOIN... ON
语法,无论如何你都应该这样做。
EXPLAIN SELECT p.media_extension,
p.media_vishash,
ANY_VALUE(a.title_url) as album_title_url,
ANY_VALUE(p.title_url) as photo_title_url,
p.media_time_created,
p.comments,
p.title,
p.text,
MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename)
AGAINST ('search query' IN BOOLEAN MODE) as score
FROM media p
STRAIGHT_JOIN album_photo_map apm ON p.media_vishash = apm.media_vishash
STRAIGHT_JOIN albums a ON apm.album_id = a.album_id
WHERE
p.level <= 5 AND
a.level <= 5 AND
MATCH (p.title,p.text,p.tag_list,p.comment_list,p.media_filename)
AGAINST ('search query' IN BOOLEAN MODE)
GROUP BY p.media_vishash
ORDER BY p.datetime_created DESC, p.media_vishash DESC
这导致更有利的优化。 它使用全文索引,然后使用索引连接到其他两个表。
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: p
partitions: NULL
type: fulltext
possible_keys: media_vishash,search
key: search
key_len: 0
ref: const
rows: 1
filtered: 100.00
Extra: Using where; Using temporary; Using filesort
*************************** 2. row ***************************
id: 1
select_type: SIMPLE
table: apm
partitions: NULL
type: index
possible_keys: PRIMARY,album_id,media_vishash
key: album_id
key_len: 130
ref: NULL
rows: 1
filtered: 100.00
Extra: Using where; Using index
*************************** 3. row ***************************
id: 1
select_type: SIMPLE
table: a
partitions: NULL
type: eq_ref
possible_keys: PRIMARY,album_id_type,level
key: PRIMARY
key_len: 130
ref: test.apm.album_id
rows: 1
filtered: 100.00
Extra: Using where
我们可以再改进一点。 由于首先读取media
表,因此需要通过album_photo_map
在专辑照片media_vishash
中进行查找。 但这不是后一个表主键的最左边一列。 如果您重新排序主键,它将能够进行更有效的连接。
alter table album_photo_map drop primary key,
add primary key(media_vishash,album_id);
PRIMARY KEY (`album_id`),
UNIQUE KEY `album_id_type` (`album_id`,`type`),
在 MySQL 中,PK 必须是唯一的并与数据聚集在一起。 因此,该唯一键是多余且不必要的。
拥有多个UNIQUE
键是不寻常的。
请为您的两个查询提供EXPLAIN SELECT...
计时查询时,请运行两次。 第二次运行可能会快很多。 这可能是由于第一次运行做了很多 I/O,而第二次发现缓存了所有必要的数据。
这里有一些尝试:
SELECT ...
FROM (your second query) AS a
JOIN album_photo_map AS apm ON apm.album_id = a.album_id
JOIN media AS p ON p.media_vishash = apm.media_vishash
WHERE ... the rest of the conditions
GROUP BY p.datetime_created, p.media_vishash
ORDER BY p.datetime_created DESC, p.media_vishash DESC
我选择使GROUP BY
与ORDER BY
BY 匹配,以消除额外的排序。
我认为,这种重写将迫使它首先使用FULLTEXT
索引,这似乎是性能的关键。 (注意EXPLAIN
是如何显示第a
被使用的。)
其他注意事项:
album_id
( album_id
)
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.