SuSE 11 sp1 x86_64 + MySQL 5.5.37
blog地址:http://blog.csdn.net/hw_libo/article/details/39583247
测试表order_line有3.2亿数据,大小约37G:
NDSC02:/data/mysql/mysql3306/data/tpcc1000 # du -shl order_line.* 12K order_line.frm 37G order_line.ibd
mysql> show table status like 'order_line'; +------------+--------+---------+------------+-----------+----------------+-------------+-----------------+--------------+-----------+-- | Name | Engine | Version | Row_format | Rows | Avg_row_length | Data_length | Max_data_length | Index_length | Data_free | A +------------+--------+---------+------------+-----------+----------------+-------------+-----------------+--------------+-----------+-- | order_line | InnoDB | 10 | Compact | 328191117 | 84 | 27771404288 | 0 | 10846420992 | 6291456 | +------------+--------+---------+------------+-----------+----------------+-------------+-----------------+--------------+-----------+-- 1 row in set (0.09 sec)
# InnoDB variables innodb_data_file_path = ibdata1:1G:autoextend innodb_buffer_pool_size = 35G innodb_file_per_table = 1 innodb_thread_concurrency = 20 innodb_flush_log_at_trx_commit = 1 innodb_log_buffer_size = 16M innodb_log_file_size = 256M innodb_log_files_in_group = 3 innodb_max_dirty_pages_pct = 50 innodb_lock_wait_timeout = 120 innodb_rollback_on_timeout innodb_status_file = 1 transaction_isolation = READ-COMMITTED
bulk_insert_buffer_size<span style="white-space:pre"> </span>= 64M
参数说明:
-e, --extended-insert,长INSERT,多row在一起批量INSERT,提高导入效率,和没有开启 -e 的备份导入耗时至少相差3、4倍,默认开启;用--extended-insert=false关闭。强烈建议开启,通过下面的测试比较就会明白为什么了。
(1)默认方式导出,也即--extended-insert=true
# time mysqldump -S /tmp/mysql.sock -uroot -proot --single-transaction -B tpcc1000 --tables order_line > ./tpcc1000_order_line1.sql real 7m38.824s user 6m44.777s sys 0m50.627s
NDSC02:/opt/mysql/backup # ls -l tpcc1000_order_line1.sql -rw-r--r-- 1 root root 24703941064 09-26 16:39 tpcc1000_order_line1.sql NDSC02:/opt/mysql/backup # du -sh tpcc1000_order_line1.sql 24G tpcc1000_order_line1.sql
# time mysqldump -S /tmp/mysql.sock -uroot -proot --single-transaction --extended-insert=false -B tpcc1000 --tables order_line > ./tpcc1000_order_line2.sql real 9m36.340s user 8m18.219s sys 1m12.241s
NDSC02:/opt/mysql/backup # ls -l tpcc1000_order_line2.sql -rw-r--r-- 1 root root 35094700366 09-26 16:49 tpcc1000_order_line2.sql NDSC02:/opt/mysql/backup # du -sh tpcc1000_order_line2.sql 33G tpcc1000_order_line2.sql可见,默认情况下(--extended-insert=true),导出37G的表,耗时7分38秒,导出文件为24G,如果关闭--extended-insert=false,同样的表,导出时耗时9分36秒,且导出文件为33G。
我测试过两次,基本一样。可以导出文件时,开启--extended-insert=true是必须的,这样导出文件小,耗时也比较少。
这里说说默认情况下(--extended-insert=true)导出的文件与使用--extended-insert=false导出的文件在导入时的性能影响。
说明:innodb_flush_log_at_trx_commit=2
这里使用了测试表orders,表的大小为2.6GB,行数为31493000行,下面是导出文件:
# du -sh tpcc1000_orders* 1.4G tpcc1000_orders1.sql ## 使用默认情况下(--extended-insert=true)导出的文件 2.3G tpcc1000_orders2.sql ## 使用--extended-insert=false导出的文件
# time mysql -f -S /tmp/mysql.sock -uroot -proot test < ./tpcc1000_orders1.sql real 12m2.184s user 0m28.538s sys 0m1.460s
# time mysql -f -S /tmp/mysql3308.sock -uroot -proot bosco2 < ./tpcc1000_orders2.sql real 276m39.231s ## 约4.6小时 user 8m13.391s sys 6m20.120s
那么使用--extended-insert=false导出表是不是一无是处呢?
并非如此。比如数据库中表中已经存在大量数据,那么再往表中导入数据时,如果出现主键数据冲突Duplicate key error,将会导致导入操作失败,但此时如果是使用--extended-insert=false导出表,导入时主键冲突的会报错Duplicate key error,但不冲突的数据仍然能正常导入。
blog地址:http://blog.csdn.net/hw_libo/article/details/39583247
-- Bosco QQ:375612082
---- END ----