问题
I want to speed up a data loading.
I use MySQL 5.5, InnoDB and have 1M rows of data (65Mb file). It takes 5 minutes.
What mysql settings and commands affect the speed of LOAD DATA INFILE for InnoDB?
Thank you.
回答1:
I can recommend these settings to improve load time:
- innodb_doublewrite = 0
- innodb_support_xa = 0
- innodb_buffer_pool_size = (50-80% of system memory)
- innodb_log_file_size = (a large number - 256M etc)
- innodb_flush_log_at_trx_commit = 0
Other than settings, there are some things you can do yourself:
- Create indexes after loading (this is a new optimization with 5.5 / InnoDB plugin).
- Sort the data file before loading.
- Split the data file, and load in parallel.
回答2:
Try removing indexes and triggers. You can re-create them after the load. Also look into using some of the high-load settings in my-huge.cnf instead of the defaults.
Some more innodb performance settings:
http://www.mysqlperformanceblog.com/2007/11/01/innodb-performance-optimization-basics/
回答3:
This might not be exactly what you're looking for but is a trick I've used in the past
ALTER TABLE TABLE_NAME DISABLE KEYS;
LOAD DATA INFILE ... ;
ALTER TABLE TABLE_NAME ENABLE KEYS;
Hope it helps.
回答4:
Also make sure that binary logging disabled if possible.
回答5:
If you are in a hurry because you are replacing the contents of a live table, then do it this way instead:
CREATE TABLE new LIKE live;
LOAD DATA ... INTO new;
RENAME TABLE live TO old, new TO live;
DROP TABLE old;
The RENAME
is 'instantaneous' and atomic, so you are 'never' down, regardless of table size.
(Hence you don't need to worry so much about speeding up the LOAD
.)
回答6:
There seems to be a high correlation between insert time and file size. I recommend partitioning the files and inserting into partition tables in parallel.
r = 0.8936003754671448
Table size growth was not correlated with insert performance (bubble size = files size).
r = 0.004109728810459718
来源:https://stackoverflow.com/questions/5020998/how-to-speed-up-a-data-loading-into-innodb-load-data-infile