I have a query below that joins 4 tables. I have added indexes and the explain output looks good with indexes used (see below).
Can I optimize the query further?
The modelXml is rather big on some records.For a big project where I get back 22 records, each with approximately 2.5 - 3MB of modelXml data, the query is taking long (total of 69MB of data returned).
I suspect this is the issue but not sure how to deal with it.
I was reading on adjusting internal mysql variables e.g key_buffer_size and table_cache. Would any of this help?
key_buffer_size is currently set at 8384512 (~8MB) and table_cache at 64
What should I increase it to?
What other variables should I be looking at to manage to speed up return of such big data?
Any other suggestions are welcome. I am a novice to mysql but really trying to get better.
SELECT `m`.`modelId`, `m`.`modelTypeId`, `m`.`modelXml`, `m`.`xmlSize`, `m`.`createdById`, `m`.`creationDate`, `m`.`modifiedDate`, `u`.`firstName`, `u`.`lastName` FROM `models_1` AS `m`
INNER JOIN `modelFolderAssociations_1` AS `mfa` ON m.modelId = mfa.modelIOId
INNER JOIN `modelFolders_1` AS `mf` ON mfa.folderId = mf.folderId
INNER JOIN `users_1` AS `u` ON m.createdById = u.userId
WHERE (m.projectId = 2) AND (mfa.folderId = 5) AND (mfa.modelIOType = 2) AND (m.modelTypeId = 2)
CREATE TABLE `models` (
`modelId` int(11) NOT NULL auto_increment,
`customerId` int(11) NOT NULL,
`groupId` int(11) NOT NULL,
`projectId` int(11) NOT NULL,
`createdById` int(11) NOT NULL,
`modelTypeId` int(11) NOT NULL,
`modelXml` longtext,
`modelSpecXml` longtext NOT NULL,
`xmlSize` bigint(20) NOT NULL default '0',
`creationDate` datetime NOT NULL,
`modifiedDate` datetime NOT NULL,
PRIMARY KEY (`modelId`,`customerId`)
) ENGINE=InnoDB AUTO_INCREMENT=75 DEFAULT CHARSET=utf8
CREATE TABLE `modelFolders` (
`folderId` int(11) NOT NULL auto_increment,
`customerId` int(11) NOT NULL,
`groupId` int(11) NOT NULL,
`projectId` int(11) NOT NULL,
`parentId` int(11) NOT NULL,
`folderName` varchar(64) NOT NULL,
`folderType` int(11) NOT NULL,
`editable` tinyint(1) NOT NULL default '1',
`nextDefaultNameNumber` int(11) NOT NULL default '1',
`creationDate` datetime NOT NULL,
`modifiedDate` datetime NOT NULL,
PRIMARY KEY (`folderId`,`customerId`),
KEY `parentId` (`parentId`)
) ENGINE=InnoDB AUTO_INCREMENT=16 DEFAULT CHARSET=utf8
CREATE TABLE `modelFolderAssociations` (
`associationId` int(11) NOT NULL auto_increment,
`customerId` int(11) NOT NULL,
`folderId` int(11) NOT NULL,
`projectId` int(11) NOT NULL,
`modelIOId` int(11) NOT NULL,
`modelIOType` tinyint(1) NOT NULL default '1',
`creationDate` datetime NOT NULL,
`modifiedDate` datetime NOT NULL,
PRIMARY KEY (`associationId`,`customerId`),
KEY `folderId` (`folderId`,`modelIOType`)
) ENGINE=InnoDB AUTO_INCREMENT=75 DEFAULT CHARSET=utf8
CREATE TABLE `users` (
`userId` int(11) NOT NULL auto_increment,
`customerId` int(11) NOT NULL,
`userName` varchar(50) NOT NULL,
`password` varchar(256) NOT NULL,
`firstName` varchar(50) default NULL,
`lastName` varchar(50) default NULL,
`creationDate` datetime NOT NULL,
`modifiedDate` datetime NOT NULL,
PRIMARY KEY (`userId`,`customerId`),
UNIQUE KEY `userName` (`userName`)
) ENGINE=InnoDB AUTO_INCREMENT=21 DEFAULT CHARSET=utf8
Explain output
+----+-------------+-------------------------+--------+---------------+----------+---------+---------------------------------------------------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------------------------+--------+---------------+----------+---------+---------------------------------------------------+------+-------------+
| 1 | SIMPLE | modelFolders | const | PRIMARY | PRIMARY | 8 | const,const | 1 | Using index |
| 1 | SIMPLE | modelFolderAssociations | ref | folderId | folderId | 5 | const,const | 22 | Using where |
| 1 | SIMPLE | models | eq_ref | PRIMARY | PRIMARY | 8 | xa_system.modelFolderAssociations.modelIOId,const | 1 | Using where |
| 1 | SIMPLE | users | eq_ref | PRIMARY | PRIMARY | 8 | xa_system.models.createdById,const | 1 | |
+----+-------------+-------------------------+--------+---------------+----------+---------+---------------------------------------------------+------+-------------+
Having large text columns, like those you're using to store XML I presume, can hurt performance regardless of how well your indexes are structured.
It can be better in these cases to move the text columns to a separate table, indexed by string length which you're already storing and CRC32.
CREATE TABLE MODEL_XML (
xmlId INT(11) unsigned NOT NULL auto_increment,
xmlSize BIGINT(20) NOT NULL default '0',
crc32 INT(11) unsigned NOT NULL,
xmlData LONGTEXT,
PRIMARY KEY (xmlId),
UNIQUE KEY (xmlSize, crc32)
)
Then the width of your columns on the table with your important indices becomes constant.
ex.
modelXmlId INT(11) unsigned NOT NULL
specXmlId INT(11) unsigned NOT NULL
It also has the benefit of being more space-efficient for redundant text (empty strings, etc) since they would all share one xmlId and thus one row in the DB.
You should index your foreign keys, this will really help with your joins:
CREATE INDEX IDX_MODELS_CUSTID
on models (customerId)
CREATE INDEX IDX_FLDR_ASSOC_MODELIO
modelFolderAssociations(modelIOId)
CREATE INDEX IDX_FLDR_ASSOC_FLDRID
modelFolderAssociations(folderId)
and so on.
Test your query after taking off the modelXml column from the SELECT clause.
If the speed is then significantly better, then the slowness comes from the volume off data to transfer, not the query itself.
Your indexes don't always match the queries and joins. If all of the columns in your WHERE and JOIN clauses are not in the index, you're forcing mysql to look at the underlying row, which will kill your performance, especially since your model rows are so wide.
For "modelFolderAssociations", you made the correct composite key for the where clause, but you should include modelIOId for the join to model.
For "models", you'd want a composite index on (modelId, projectId, modelTypeId, createdById ) to cover the incoming link from mfa, the two items in the where clause, and the outbound link to users.
For "modelFolders" and "users", you have the incoming join covered in your primary keys.
The engine is only going to use one index, so adding extra index single indexes (as Mike suggested) won't be as good.
Related
I have some trouble with the execution time of a script. The query takes a lot of time. This is the query:
select avg(price) from voiture where duration<30 AND make="Audi" AND model="A4"
+---+---+---+---+---+---+---+---+---+---+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+---+---+---+---+---+---+---+---+---+---+
| 1 | SIMPLE | voiture | ALL | NULL | NULL | NULL | NULL | 1376949 | Using where |
+---+---+---+---+---+---+---+---+---+---+
select price from voiture where duration<30 AND make="Audi" AND model="A4"
+---+---+---+---+---+---+---+---+---+---+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+---+---+---+---+---+---+---+---+---+---+
| 1 | SIMPLE | voiture | ALL | NULL | NULL | NULL | NULL | 1376949 | Using where |
+---+---+---+---+---+---+---+---+---+---+
This query take around 2 seconds to be executed on the phpMyAdmin interface. I tried to see what the issue was and removing the avg function makes the query lasts around 0.0080 seconds.
I asked myself how long it would take to make the calculate the avg in the php script, but the query with or withouth avg takes around 2 seconds both.
So I decided to take all the values of my table and make the process in the script, so I use this query:
select * from voiture where duration<30 AND make="Audi" AND model="A4"
+---+---+---+---+---+---+---+---+---+---+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+---+---+---+---+---+---+---+---+---+---+
| 1 | SIMPLE | voiture | ALL | NULL | NULL | NULL | NULL | 1376949 | Using where |
+---+---+---+---+---+---+---+---+---+---+
On the phpMyAdmin interface, it takes 0.0112 seconds. But in my php script, it takes 25 seconds !
$timestart2=microtime(true);
$querysec='select * from voiture where duration<30 AND make="Audi" AND model="A4"';
$requestsec=mysql_query($querysec) or die(mysql_error());
$timeend2=microtime(true);
$time2=$timeend2-$timestart2;
$page_load_time2 = number_format($time2, 9);
Here is the table structure:
CREATE TABLE `voiture` (
`carid` bigint(9) NOT NULL,
`serviceid` bigint(9) NOT NULL,
`service` varchar(256) NOT NULL,
`model` varchar(256) DEFAULT NULL,
`gearingType` varchar(256) DEFAULT NULL,
`displacement` int(5) DEFAULT NULL,
`cylinders` int(2) DEFAULT NULL,
`fuel` varchar(32) DEFAULT NULL,
`mileage` int(7) DEFAULT NULL,
`existFlag` tinyint(1) DEFAULT NULL,
`lastUpdate` date DEFAULT NULL,
`version` varchar(256) DEFAULT NULL,
`bodyType` varchar(256) DEFAULT NULL,
`firstRegistration` date DEFAULT NULL,
`powerHp` int(4) DEFAULT NULL,
`powerKw` int(4) DEFAULT NULL,
`vat` varchar(256) DEFAULT NULL,
`price` decimal(12,2) DEFAULT NULL,
`duration` int(3) DEFAULT NULL,
`pageUrl` varchar(256) DEFAULT NULL,
`carImg` varchar(256) DEFAULT NULL,
`color` varchar(256) DEFAULT NULL,
`doors` int(1) DEFAULT NULL,
`seats` int(1) DEFAULT NULL,
`prevOwner` int(1) DEFAULT NULL,
`co2` varchar(256) DEFAULT NULL,
`consumption` varchar(256) DEFAULT NULL,
`gears` int(1) DEFAULT NULL,
`equipment` varchar(1024) NOT NULL,
`make` varchar(256) NOT NULL,
`country` varchar(3) NOT NULL
)
There's an index on carid and serviceid
Why does my query takes so long to be executed ? Is there a way it can be improved ?
Why is the execution time different from phpMyAdmin and my php script ?
On the phpMyAdmin interface, it takes 0.0112 seconds. But in my php
script, it takes 25 seconds!
phpMyAdmin interface adds LIMIT to each query. By default it's LIMIT 30.
To decrease time of your aggregate query you need to create indexes for each condition you use(or one composite index, may be).
So, try to create indexes for your model, make and duration fields.
Also, your table is too denormalized. You can to create pair of table to normalize it for a bit.
Ex: Vendors(id, name), Models(id, name) and modify your voiture to have vendor_id/model_id fields instead of text make/model.
Then your initial query will look like:
select avg(t.price) from voiture t
INNER JOIN Models m ON m.id = t.model_id
INNER JOIN Vendors v ON v.id = t.vendor_id
where t.duration<30 AND v.name="Audi" AND m.name="A4"
It will scan light lookup tables for text matches and operate with your heavy table with indexed ids.
There are many possible solution, first thing you can do is create index at DB level this can also improve your execution time.
second this check server, there may be some processes which would occupying your server resources, and making your server slow.
You can make it faster using GROUP BY like
select AVG(price) from voiture where duration<30 AND make="Audi" AND model="A4" GROUP BY make
Also you can add an index for make column
ALTER TABLE `voiture` ADD INDEX `make` (`make`)
I have below query:
SELECT bank_code, bank_name
FROM system_bank_info
WHERE company_id=1 AND country_id=103 AND status='ACTIVE'
GROUP BY bank_name
ORDER BY bank_name ASC
TABLE SCHEMA:
CREATE TABLE `system_bank_info` (
`id` int(11) NOT NULL AUTO_INCREMENT,
`company_id` int(11) NOT NULL,
`country_id` int(11) NOT NULL,
`bank_code` varchar(50) NOT NULL,
`bank_name` varchar(255) NOT NULL,
`branch_code` varchar(50) NOT NULL,
`branch_name` varchar(255) NOT NULL,
`status` enum('ACTIVE','INACTIVE') NOT NULL DEFAULT 'ACTIVE',
`bank_state` varchar(255) DEFAULT NULL,
`bank_district` varchar(255) DEFAULT NULL,
`bank_city` varchar(255) DEFAULT NULL,
`bank_address` text,
`bank_contact` varchar(255) DEFAULT NULL,
`service_type` enum('INSTA_CREDIT') DEFAULT NULL,
PRIMARY KEY (`id`),
KEY `country_id` (`country_id`),
KEY `bank_code` (`bank_code`),
KEY `branch_code` (`branch_code`),
KEY `company_id_country_id_status` (`company_id`,`country_id`,`status`),
CONSTRAINT `system_bank_info_ibfk_1` FOREIGN KEY (`company_id`) REFERENCES `company` (`id`),
CONSTRAINT `system_bank_info_ibfk_2` FOREIGN KEY (`country_id`) REFERENCES `system_country_list` (`country_id`)
) ENGINE=InnoDB AUTO_INCREMENT=150398 DEFAULT CHARSET=utf8
EXPLAIN:
+----+-------------+------------------+------+-----------------------------------------+------------+---------+-------+-------+----------------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+------------------+------+-----------------------------------------+------------+---------+-------+-------+----------------------------------------------+
| 1 | SIMPLE | system_bank_info | ref | country_id,company_id_country_id_status | country_id | 4 | const | 59324 | Using where; Using temporary; Using filesort |
+----+-------------+------------------+------+-----------------------------------------+------------+---------+-------+-------+----------------------------------------------+
As you can see there is an index called company_id_country_id_status. But, query still uses country_id only. How can I make this query more efficient?
I also tried adding indexes bank_code and bank_name. But, same result!
You can give hints to the optimizer so that it uses a specific index. USE INDEX ()
SELECT bank_code, bank_name
FROM system_bank_info USE INDEX (company_id_country_id_status)
WHERE company_id=1 AND country_id=103 AND status='ACTIVE'
I have two tables in my mysql database called forum_topics and forum_replies.
I am looking to have an overview of the latest posts / replies added, like this: http://prntscr.com/6ixtz4
to do so, i need a way to makee it sort by the time in both tables, and if a forum_reply is in the result it need to get its topic from the forum_topics.
How can i make this work? I have no mysql query to referer to to make this work.
CREATE TABLE IF NOT EXISTS `forum_topics` (
`id` int(255) NOT NULL AUTO_INCREMENT,
`topic` varchar(255) NOT NULL,
`thread` varchar(255) NOT NULL,
`level` int(2) NOT NULL DEFAULT '0',
`creator` int(255) NOT NULL,
`time` varchar(255) NOT NULL,
`innlegg` text NOT NULL,
`ip` varchar(255) NOT NULL,
`locked` enum('yes','no') NOT NULL DEFAULT 'no',
`deleted` enum('yes','no') NOT NULL DEFAULT 'no',
PRIMARY KEY (`id`),
UNIQUE KEY `id` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=5 ;
CREATE TABLE IF NOT EXISTS `forum_replies` (
`id` int(255) NOT NULL AUTO_INCREMENT,
`topicid` int(255) NOT NULL,
`userid` int(255) NOT NULL,
`time` int(255) NOT NULL,
`reply` text NOT NULL,
`deleted` enum('yes','no') NOT NULL DEFAULT 'no',
PRIMARY KEY (`id`),
UNIQUE KEY `id` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=20 ;
exampledata:
in table forum_topics
id |topic| time|
1 | test | 10
2 | test2 | 29
in table forum replies:
id | topicid | time |
1 | 1 | 18
2 | 2 | 28
As in this example i would like the outcome to be sorted something like this:
10
18
28
29
if forum_replies is one of the results it would need some data from forum_topics, but if forum_topics is one of the results it wouldnt need any data from the forum_replies.
The topicid is the id from the forum_topics of the topic that the reply is a response to.
So in the end i want to create the example output as in the image.
EDIT:
it would be a result containing in following order:
10
18
28
29
By loading the data in the example data sets, this will get the time from both tables, but sort the time from both as if they were in one table. The below query gets the results, and then orders them in order of time. Since both tables have the same structure, you can "union" them together as if they were one table and sort by time using "order_by"
select time
from forum_topics
UNION
select time
from forum_replies
order by time
Note: I was able to load both tables in the above request but without the last line on each. In other words I trimmed the part referring to the engine, and then just created the tables, and then entered the data provided..
I'm sorry I have to say the table structure of a forum post is incorrect. In your case, you will be very hard to sort a post based on topic and reply. I would suggest a table structure like this:
CREATE TABLE IF NOT EXISTS `forum_topics` (
`tid` int(255) NOT NULL AUTO_INCREMENT, // Topic ID
`tsubject` varchar(255) NOT NULL, // Topic Subject
`level` int(2) NOT NULL DEFAULT '0',
`creator` int(255) NOT NULL,
`innlegg` text NOT NULL, // I don't know what this means ...
`locked` enum('yes','no') NOT NULL DEFAULT 'no',
`deleted` enum('yes','no') NOT NULL DEFAULT 'no',
PRIMARY KEY (`tid`),
UNIQUE KEY `id` (`tid`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=5;
CREATE TABLE IF NOT EXISTS `forum_posts` (
`pid` int(255) NOT NULL AUTO_INCREMENT,
`tid` int(255) NOT NULL,
`userid` int(255) NOT NULL,
`time` int(255) NOT NULL,
`pcontent` text NOT NULL, // Post Content
`isauthorpost` int(1) NOT NULL, // Optional: If this is the first post of the topic
`deleted` enum('yes','no') NOT NULL DEFAULT 'no',
PRIMARY KEY (`pid`),
UNIQUE KEY `id` (`pid`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=20 ;
So now you will have a topic id to follow by the first. And you can just search for all replies with ORDER BY pid, for the topic details, you can use LEFT JOIN forum_topics or INNER JOIN to finish.
Table definition and queries explained:
item |
CREATE TABLE `item` (
`item_id` int(11) NOT NULL AUTO_INCREMENT,
`item_type_id` int(11) NOT NULL,
`brand_id` int(11) NOT NULL,
`site_id` int(11) NOT NULL,
`seller_id` int(11) NOT NULL,
`title` varchar(175) NOT NULL,
`desc` text NOT NULL,
`url` varchar(767) NOT NULL,
`price` int(11) NOT NULL,
`photo` varchar(255) NOT NULL,
`photo_file` varchar(255) NOT NULL,
`photo_type` varchar(32) NOT NULL,
`has_photo` enum('yes','no','pending') NOT NULL DEFAULT 'pending',
`added_at` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`updated_at` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
`created_at` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
`normalized_time` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',
`location` varchar(128) NOT NULL,
PRIMARY KEY (`item_id`),
KEY `item_type_id` (`item_type_id`),
KEY `brand_id` (`brand_id`),
KEY `site_id` (`site_id`),
KEY `seller_id` (`seller_id`),
KEY `created_at` (`created_at`),
KEY `added_at` (`added_at`),
KEY `normalized_time` (`normalized_time`),
KEY `typephototime` (`item_type_id`,`has_photo`,`normalized_time`),
KEY `brandidphoto` (`brand_id`,`item_type_id`,`has_photo`),
KEY `brandidphoto2` (`brand_id`,`item_type_id`,`has_photo`),
KEY `idphoto` (`item_type_id`,`has_photo`),
KEY `idphototime` (`item_type_id`,`has_photo`,`normalized_time`),
KEY `idphoto2` (`item_type_id`,`has_photo`),
KEY `typepricebrandid` (`item_type_id`,`price`,`brand_id`,`item_id`),
KEY `sellertypephototime` (`seller_id`,`item_type_id`,`has_photo`,`normalized_time`),
KEY `typephoto` (`item_type_id`,`has_photo`)
) ENGINE=MyISAM AUTO_INCREMENT=508885 DEFAULT CHARSET=latin1 |
mysql> explain SELECT item.* FROM item WHERE item.item_type_id = "1" AND item.has_photo = "yes" ORDER BY normalized_time DESC LIMIT 1;
+----+-------------+-------+------+------------------------------------------------------------------------------------+---------------+---------+-------------+-------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+------+------------------------------------------------------------------------------------+---------------+---------+-------------+-------+-------------+
| 1 | SIMPLE | item | ref | item_type_id,typephototime,idphoto,idphototime,idphoto2,typepricebrandid,typephoto | typephototime | 5 | const,const | 69528 | Using where |
+----+-------------+-------+------+------------------------------------------------------------------------------------+---------------+---------+-------------+-------+-------------+
1 row in set (0.02 sec)
mysql> explain SELECT * FROM item WHERE item_type_id = "1" AND (price BETWEEN "25" AND "275") AND brand_id = "10" ORDER BY item_id DESC LIMIT 1;
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+---------+---------+------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+---------+---------+------+------+-------------+
| 1 | SIMPLE | item | index | item_type_id,brand_id,typephototime,brandidphoto,brandidphoto2,idphoto,idphototime,idphoto2,typepricebrandid,typephoto | PRIMARY | 4 | NULL | 203 | Using where |
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+---------+---------+------+------+-------------+
1 row in set (0.01 sec)
mysql> explain SELECT item.* FROM item WHERE item.brand_id = "10" AND item.item_type_id = "1" AND item.has_photo = "yes" ORDER BY normalized_time DESC LIMIT 1;
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+-----------------+---------+------+--------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+-----------------+---------+------+--------+-------------+
| 1 | SIMPLE | item | index | item_type_id,brand_id,typephototime,brandidphoto,brandidphoto2,idphoto,idphototime,idphoto2,typepricebrandid,typephoto | normalized_time | 8 | NULL | 502397 | Using where |
+----+-------------+-------+-------+------------------------------------------------------------------------------------------------------------------------+-----------------+---------+------+--------+-------------+
1 row in set (2.15 sec)
mysql> explain SELECT COUNT(*) FROM item WHERE item.item_type_id = "1" AND item.has_photo = "yes" ;
+----+-------------+-------+------+------------------------------------------------------------------------------------+-----------+---------+-------------+-------+--------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+------+------------------------------------------------------------------------------------+-----------+---------+-------------+-------+--------------------------+
| 1 | SIMPLE | item | ref | item_type_id,typephototime,idphoto,idphototime,idphoto2,typepricebrandid,typephoto | typephoto | 5 | const,const | 71135 | Using where; Using index |
+----+-------------+-------+------+------------------------------------------------------------------------------------+-----------+---------+-------------+-------+--------------------------+
1 row in set (0.01 sec)
The following indexes are redundant because they match the left columns of another index. You can almost certainly drop these indexes and save some space and overhead.
KEY `item_type_id` (`item_type_id`), /* redundant */
KEY `brand_id` (`brand_id`), /* redundant */
KEY `seller_id` (`seller_id`), /* redundant */
KEY `idphototime` (`item_type_id`,`has_photo`,`normalized_time`), /* redundant */
KEY `brandidphoto2` (`brand_id`,`item_type_id`,`has_photo`), /* redundant */
KEY `idphoto` (`item_type_id`,`has_photo`), /* redundant */
KEY `idphoto2` (`item_type_id`,`has_photo`), /* redundant */
KEY `typephoto` (`item_type_id`,`has_photo`) /* redundant */
That leaves the following indexes:
KEY `site_id` (`site_id`),
KEY `created_at` (`created_at`),
KEY `added_at` (`added_at`),
KEY `normalized_time` (`normalized_time`),
KEY `brandidphoto` (`brand_id`,`item_type_id`,`has_photo`),
KEY `typephototime` (`item_type_id`,`has_photo`,`normalized_time`),
KEY `typepricebrandid` (`item_type_id`,`price`,`brand_id`,`item_id`),
KEY `sellertypephototime` (`seller_id`,`item_type_id`,`has_photo`,`normalized_time`),
You can also use a tool like pt-duplicate-key-checker to find redundant indexes.
Next consider the storage engine:
) ENGINE=MyISAM AUTO_INCREMENT=508885 DEFAULT CHARSET=latin1;
Almost always, InnoDB is a better choice than MyISAM. Not only for performance, but for data integrity and crash safety. InnoDB has been the default storage engine since 2010, and it's the only storage engine that is actively getting improved. I'd recommend making a copy of this table, changing the storage engine to InnoDB, and compare its performance with respect to your queries.
Next let's consider indexes for the queries:
SELECT item.* FROM `item` WHERE item.item_type_id = "1" AND item.has_photo = "yes"
ORDER BY normalized_time DESC LIMIT 1;
I would choose an index on (item_type_id, has_photo, normalized_time) and that's the index it's currently using, which is typephototime.
One way to optimize this further would be to fetch only the columns in the index. That's when you see "Using index" in the EXPLAIN plan, it can be a huge improvement for performance.
Another important factor is to make sure that your index is cached in memory: increase key_buffer_size if you use MyISAM or innodb_buffer_pool_size if you use InnoDB to be as large as all the indexes you want to remain in memory. Because you don't want to run a query that needs to scan an index larger than your buffers; it causes a lot of swapping.
SELECT * FROM `item` WHERE item_type_id = "1" AND (price BETWEEN "25" AND "275") AND brand_id = "10"
ORDER BY item_id DESC LIMIT 1;
I would choose an index on (item_type_id, brand_id, price), but this query is currently using the PRIMARY index. You should create a new index.
SELECT item.* FROM `item` WHERE item.brand_id = "10" AND item.item_type_id = "1" AND item.has_photo = "yes"
ORDER BY normalized_time DESC LIMIT 1;
I would choose an index on (item_type_id, brand_id, has_photo, normalized_time). You should create a new index.
SELECT COUNT(*) FROM `item` WHERE item.item_type_id = "1" AND item.has_photo = "yes" ;
I would choose an index on (item_type_id, has_photo) and that's the index it's currently using, which is typephoto. It's also getting the "Using index" optimization, so the only other improvement could be to make sure there's enough buffer to hold the index in memory.
It's hard to optimize SELECT COUNT(*) queries because they naturally have to scan a lot of rows.
The other strategy to optimize COUNT(*) is to calculate the counts offline, and store them either in a summary table or in an in-memory cache like memcached so you don't have to recalculate them every time someone loads a page. But that means you have to update the counts every time someone adds or deletes a row in the item table, which could be more costly depending on how frequently that happens.
A few things I would suggest changing:
You don't need all those indexes. You really only need indexes on fields that are accessed a lot, like foreign key fields. Remove all the indexes except ones on ID fields.
You should be storing dates as nulls unless there is actual data.
Stay away from the enum data type, use smallint with flags representing each value. Example, 0 pending, 1 yes, 2 no.
Alongside reducing the size of the database, it makes things much cleaner. Your new table structure would look like so:
CREATE TABLE `item` (
`item_id` int(11) NOT NULL AUTO_INCREMENT,
`item_type_id` int(11) NOT NULL,
`brand_id` int(11) NOT NULL,
`site_id` int(11) NOT NULL,
`seller_id` int(11) NOT NULL,
`title` varchar(175) NOT NULL,
`desc` text NOT NULL,
`url` varchar(767) NOT NULL,
`price` int(11) NOT NULL,
`photo` varchar(255) NOT NULL,
`photo_file` varchar(255) NOT NULL,
`photo_type` varchar(32) NOT NULL,
`has_photo` smallint NOT NULL DEFAULT 0,
`added_at` timestamp NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`updated_at` datetime NULL DEFAULT NULL,
`created_at` datetime NULL DEFAULT NULL,
`normalized_time` datetime NULL DEFAULT NULL,
`location` varchar(128) NULL,
PRIMARY KEY (`item_id`),
KEY `item_type_id` (`item_type_id`),
KEY `brand_id` (`brand_id`),
KEY `site_id` (`site_id`),
KEY `seller_id` (`seller_id`)
) ENGINE=MyISAM AUTO_INCREMENT=508885 DEFAULT CHARSET=latin1;
I would also suggest using utf8_unicode_ci as the collate, utf8 as the charset, and InnoDB as the engine.
But first off, remove all those keys and try again. Also remove the aliasing on the 3rd query.
SELECT * FROM item WHERE brand_id = "10" AND item_type_id = "1" AND has_photo = "yes" ORDER BY normalized_time DESC LIMIT 1;
I am using the below SQL query but it takes more than 180 sec to execute. Is there a way to speed it up ? This SQL give me the pic_id and of all the females.
SELECT pic_id, small
FROM picture
WHERE hide =0
AND userhide =0
AND `fbid`
IN (
SELECT fbid
FROM user
WHERE gender = "female"
)
ORDER BY `picture`.`pic_id` ASC
LIMIT 1500 , 200
The Explain SQL
id select_type table type possible_keys key key_len ref rows Extra
1 PRIMARY picture index NULL PRIMARY 4 NULL 1700 Using where
2 DEPENDENT SUBQUERY user ALL NULL NULL NULL NULL 7496 Using where
--- Result of explain statement for Tim's sql answer --
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE picture ALL NULL NULL NULL NULL 41443 Using where; Using temporary; Using filesort
1 SIMPLE user ALL NULL NULL NULL NULL 7501 Using where; Using join buffer
-- Structure ---
CREATE TABLE `user` (
`sid` int(11) NOT NULL AUTO_INCREMENT,
`name` varchar(255) CHARACTER SET utf8 NOT NULL,
`first_name` varchar(255) CHARACTER SET utf8 NOT NULL,
`username` varchar(255) CHARACTER SET utf8 NOT NULL,
`birthday` date NOT NULL,
`location` varchar(255) CHARACTER SET utf8 NOT NULL,
`gender` varchar(6) CHARACTER SET utf8 NOT NULL,
`created` date NOT NULL,
`fbid` bigint(50) NOT NULL,
`token` varchar(255) CHARACTER SET utf8 NOT NULL,
`relationship_status` varchar(20) CHARACTER SET utf8 NOT NULL,
`smallest` varchar(255) CHARACTER SET utf8 NOT NULL,
`email` varchar(40) CHARACTER SET utf8 NOT NULL,
`ref` varchar(15) NOT NULL,
PRIMARY KEY (`sid`),
KEY `gender` (`gender`),
KEY `fbid` (`fbid`)
) ENGINE=MyISAM AUTO_INCREMENT=7595 DEFAULT CHARSET=latin
---- structure of picture table ---
CREATE TABLE `picture` (
`fbid` bigint(50) NOT NULL,
`pic_id` int(11) NOT NULL AUTO_INCREMENT,
`pic_location` varchar(255) NOT NULL,
`hide` int(1) NOT NULL,
`small` varchar(255) NOT NULL,
`userhide` int(1) NOT NULL,
`likes` int(10) NOT NULL,
`hot` int(1) NOT NULL,
PRIMARY KEY (`pic_id`),
UNIQUE KEY `pic_location` (`pic_location`),
UNIQUE KEY `small` (`small`),
KEY `fbid` (`fbid`),
KEY `hide` (`hide`),
KEY `userhide` (`userhide`)
) ENGINE=MyISAM AUTO_INCREMENT=42749 DEFAULT CHARSET=latin1
try something like this:
SELECT pic_id, small
FROM picture
INNER JOIN user ON ( picture.fbid = user.fbid and user.gender='female' )
WHERE hide =0
AND userhide =0
ORDER BY `picture`.`pic_id` ASC
LIMIT 1500 , 200
I put gender in the join because a query will not return rows that don't have a match on an inner join.
You should also read this stack overflow topic
EDIT:
make sure you have indexed the following fields:
picture.fbid
user.fbid
user.gender
picture.hide
picture.userhide
Try using a join instead:
SELECT p.pic_id, p.small
FROM picture p
INNER JOIN fbid f USING ( fbid )
WHERE p.hide =0
AND p.userhide =0
AND f.gender = 'female'
ORDER BY `picture`.`pic_id` ASC
LIMIT 1500 , 200