Using MySQL 5.6 with InnoDB storage engine for most of the tables. InnoDB buffer pool size is 15 GB and Innodb DB + indexes are around 10 GB. Server has 32GB RAM and is running Cent OS 7 x64.

I have one big table which contains around 10 millions + records.

I get an updated dump file from a remote server every 24 hours. The file is in csv format. I don't have control over that format. The file is ~750 MB. I tried inserting data to a MyISAM table row by row and it took 35 minutes.

I need to take only 3 values per line out of 10-12 from the file and update it in the database.

What's the best way to achieve something like this ?

I need to do this daily.

Currently Flow is like this:

1. mysqli_begin_transaction


2. Read Dump file line by line


3. Update each record Line by Line.


4. mysqli_commit

Above operations takes around 30-40 minutes to complete and while doing this, there are other updates going on which gives me

Lock wait timeout exceeded; try restarting transaction

Update 1

data loading in new table using LOAD DATA LOCAL INFILE. In MyISAM it took 38.93 sec while in InnoDB it took 7 min 5.21 sec. Then I did:

UPDATE table1 t1, table2 t2

SET 

t1.field1 = t2.field1,

t1.field2 = t2.field2,

t1.field3 = t2.field3

WHERE t1.field10 = t2.field10



Query OK, 434914 rows affected (22 hours 14 min 47.55 sec)

Update 2

same update with join query

UPDATE table1 a JOIN table2 b 

ON a.field1 = b.field1 

SET 

a.field2 = b.field2,

a.field3 = b.field3,

a.field4 = b.field4



(14 hours 56 min 46.85 sec)

Clarifications from questions in comments:

• About 6% of the rows in the table will be updated by the file, but sometimes it can be as much as 25%.


• There are indexes on the fields being updated. There are 12 indexes on the table, and 8 indexes include the update fields.


• It is not necessary to do the update in one transaction. It can take time but not more than 24 hours. I am looking to get it done in 1 hour without locking the whole table, as later I have to update the sphinx index which is dependent on this table. It does not matter if the steps take a longer duration as long as the database is available for other tasks.


• I could modify the csv format in a preprocess step. The only thing that matters is quick update and without locking.


• Table 2 is MyISAM. It is the newly created table from csv file using load data infile. MYI file size is 452 MB. Table 2 is indexed on the field1 column.


• MYD of the MyISAM table is 663MB.

Update 3 :

here are more details about both table.

CREATE TABLE `content` (

  `hash` char(40) CHARACTER SET ascii NOT NULL DEFAULT '',

  `title` varchar(255) COLLATE utf8_unicode_ci NOT NULL DEFAULT '',

  `og_name` varchar(255) COLLATE utf8_unicode_ci NOT NULL DEFAULT '',

  `keywords` varchar(255) COLLATE utf8_unicode_ci NOT NULL DEFAULT '',

  `files_count` smallint(5) unsigned NOT NULL DEFAULT '0',

  `more_files` smallint(5) unsigned NOT NULL DEFAULT '0',

  `files` varchar(255) COLLATE utf8_unicode_ci NOT NULL DEFAULT '0',

  `category` smallint(3) unsigned NOT NULL DEFAULT '600',

  `size` bigint(19) unsigned NOT NULL DEFAULT '0',

  `downloaders` int(11) NOT NULL DEFAULT '0',

  `completed` int(11) NOT NULL DEFAULT '0',

  `uploaders` int(11) NOT NULL DEFAULT '0',

  `creation_date` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',

  `upload_date` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',

  `last_updated` datetime NOT NULL DEFAULT '0000-00-00 00:00:00',

  `vote_up` int(11) unsigned NOT NULL DEFAULT '0',

  `vote_down` int(11) unsigned NOT NULL DEFAULT '0',

  `comments_count` int(11) NOT NULL DEFAULT '0',

  `imdb` int(8) unsigned NOT NULL DEFAULT '0',

  `video_sample` tinyint(1) NOT NULL DEFAULT '0',

  `video_quality` tinyint(2) NOT NULL DEFAULT '0',

  `audio_lang` varchar(127) CHARACTER SET ascii NOT NULL DEFAULT '',

  `subtitle_lang` varchar(127) CHARACTER SET ascii NOT NULL DEFAULT '',

  `verified` tinyint(1) unsigned NOT NULL DEFAULT '0',

  `uploader` int(11) unsigned NOT NULL DEFAULT '0',

  `anonymous` tinyint(1) NOT NULL DEFAULT '0',

  `enabled` tinyint(1) unsigned NOT NULL DEFAULT '0',

  `tfile_size` int(11) unsigned NOT NULL DEFAULT '0',

  `scrape_source` tinyint(1) unsigned NOT NULL DEFAULT '0',

  `record_num` int(11) unsigned NOT NULL AUTO_INCREMENT,

  PRIMARY KEY (`record_num`),

  UNIQUE KEY `hash` (`hash`),

  KEY `uploaders` (`uploaders`),

  KEY `tfile_size` (`tfile_size`),

  KEY `enabled_category_upload_date_verified_` (`enabled`,`category`,`upload_date`,`verified`),

  KEY `enabled_upload_date_verified_` (`enabled`,`upload_date`,`verified`),

  KEY `enabled_category_verified_` (`enabled`,`category`,`verified`),

  KEY `enabled_verified_` (`enabled`,`verified`),

  KEY `enabled_uploader_` (`enabled`,`uploader`),

  KEY `anonymous_uploader_` (`anonymous`,`uploader`),

  KEY `enabled_uploaders_upload_date_` (`enabled`,`uploaders`,`upload_date`),

  KEY `enabled_verified_category` (`enabled`,`verified`,`category`),

  KEY `verified_enabled_category` (`verified`,`enabled`,`category`)

) ENGINE=InnoDB AUTO_INCREMENT=7551163 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci ROW_FORMAT=FIXED





CREATE TABLE `content_csv_dump_temp` (

  `hash` char(40) CHARACTER SET ascii NOT NULL DEFAULT '',

  `title` varchar(255) COLLATE utf8_unicode_ci NOT NULL,

  `category_id` int(11) unsigned NOT NULL DEFAULT '0',

  `uploaders` int(11) unsigned NOT NULL DEFAULT '0',

  `downloaders` int(11) unsigned NOT NULL DEFAULT '0',

  `verified` tinyint(1) unsigned NOT NULL DEFAULT '0',

  PRIMARY KEY (`hash`)

) ENGINE=MyISAM DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci

and here is the update query which updates content table using data from content_csv_dump_temp

UPDATE content a JOIN content_csv_dump_temp b 

ON a.hash = b.hash 

SET 

a.uploaders = b.uploaders,

a.downloaders = b.downloaders,

a.verified = b.verified

update 4:

all above testing was done on test machine., but now i did same tests on the production machine, and queries are very fast.

mysql> UPDATE content_test a JOIN content_csv_dump_temp b

    -> ON a.hash = b.hash

    -> SET

    -> a.uploaders = b.uploaders,

    -> a.downloaders = b.downloaders,

    -> a.verified = b.verified;

Query OK, 2673528 rows affected (7 min 50.42 sec)

Rows matched: 7044818  Changed: 2673528  Warnings: 0

i apologize for my mistake. Its better to use join instead of each record update. now i am trying to improve mpre using the index suggested by rick_james , will update once bench-marking is done.

Based on my experience, I would use LOAD DATA INFILE to import your CSV File.

The LOAD DATA INFILE statement reads rows from a text file into a
table at a very high speed.

Example I found on the internet Load Data example. I tested this example on my box and worked fine

Example Table

CREATE TABLE example (

  `Id` int(11) NOT NULL AUTO_INCREMENT,

  `Column2` varchar(14) NOT NULL,

  `Column3` varchar(14) NOT NULL,

  `Column4` varchar(14) NOT NULL,

  `Column5` DATE NOT NULL,

  PRIMARY KEY (`Id`)

) ENGINE=InnoDB

Example CSV File

# more /tmp/example.csv

Column1,Column2,Column3,Column4,Column5

1,A,Foo,sdsdsd,4/13/2013

2,B,Bar,sdsa,4/12/2013

3,C,Foo,wewqe,3/12/2013

4,D,Bar,asdsad,2/1/2013

5,E,FOObar,wewqe,5/1/2013

Import Statement to be run from MySQL console

LOAD DATA LOCAL INFILE '/tmp/example.csv'

    -> INTO TABLE example

    -> FIELDS TERMINATED BY ','

    -> LINES TERMINATED BY 'n'

    -> IGNORE 1 LINES

    -> (id, Column3,Column4, @Column5)

    -> set

    -> Column5 = str_to_date(@Column5, '%m/%d/%Y');

Result

MySQL [testcsv]> select * from example;

+----+---------+---------+---------+------------+

| Id | Column2 | Column3 | Column4 | Column5    |

+----+---------+---------+---------+------------+

|  1 |         | Column2 | Column3 | 0000-00-00 |

|  2 |         | B       | Bar     | 0000-00-00 |

|  3 |         | C       | Foo     | 0000-00-00 |

|  4 |         | D       | Bar     | 0000-00-00 |

|  5 |         | E       | FOObar  | 0000-00-00 |

+----+---------+---------+---------+------------+

IGNORE just simply ignores the first line which are column headers.

After IGNORE, we are specifying the columns (skipping column2), to import, which matches one of the criteria in your question.

Here is another example directly from Oracle : LOAD DATA INFILE example

This should be enough to get you started.

In light of all the things mentioned, it looks like the bottleneck is the join itself.

ASPECT #1 : Join Buffer Size

In all likelihood, your join_buffer_size is probably too low.

According to the MySQL Documentation on How MySQL Uses the Join Buffer Cache

We only store the used columns in the join buffer, not the whole rows.

This being the case, make the keys of the join buffer stay in RAM.

You have 10 million rows times 4 bytes for each key. That's about 40M.

Try bumping it up in the session to 42M (a little bigger than 40M)

SET join_buffer_size = 1024 * 1024 * 42;

UPDATE table1 a JOIN table2 b 

ON a.field1 = b.field1 

SET 

a.field2 = b.field2,

a.field3 = b.field3,

a.field4 = b.field4;

If this does the trick, proceed to add this to my.cnf

[mysqld]

join_buffer_size = 42M

Restarting mysqld not required for new connections. Just run

mysql> SET GLOBAL join_buffer_size = 1024 * 1024 * 42;

ASPECT #2 : Join Operation

You could manipulate the style of the join operation by tweeking the optimizer

According to MySQL Documentation on Block Nested-Loop and Batched Key Access Joins

When BKA is used, the value of join_buffer_size defines how large the batch of keys is in each request to the storage engine. The larger the buffer, the more sequential access will be to the right hand table of a join operation, which can significantly improve performance.

For BKA to be used, the batched_key_access flag of the optimizer_switch system variable must be set to on. BKA uses MRR, so the mrr flag must also be on. Currently, the cost estimation for MRR is too pessimistic. Hence, it is also necessary for mrr_cost_based to be off for BKA to be used.

This same page recommends doing this:

mysql> SET optimizer_switch='mrr=on,mrr_cost_based=off,batched_key_access=on';

ASPECT #3 : Writing Updates to Disk (OPTIONAL)

Most forget to increase the innodb_write_io_threads to write dirty pages out of the buffer pool faster.

[mysqld]

innodb_write_io_threads = 16

You will have to restart MySQL for this change

GIVE IT A TRY !!!

1. 
   CREATE TABLE that matches the CSV


2. 
   LOAD DATA into that table


3. UPDATE real_table JOIN csv_table ON ... SET ..., ..., ...;


4. DROP TABLE csv_table;

Step 3 will be a lot faster than row-by-row, but it will still lock all the rows in the table for a non-trivial amount of time. If this lock time is more important than how long the entire process takes then, ...

If nothing else is writing to the table, then...

1. 
   CREATE TABLE that matches the CSV; no indexes except what is needed in the JOIN in the UPDATE. If unique, make it PRIMARY KEY.


2. 
   LOAD DATA into that table


3. copy the real_table to new_table (CREATE ... SELECT)


4. UPDATE new_table JOIN csv_table ON ... SET ..., ..., ...;


5. RENAME TABLE real_table TO old, new_table TO real_table;


6. DROP TABLE csv_table, old;

Step 3 is faster than the update, especially if unnecessary indexes are left off.

Step 5 is "instantaneous".