When messing with MySQL's filesystem you have to stop the MySQL server. To avoid downtime on your live machine, use a backup/virtualised machine with the SAME VERSION of MySQL server. While the BACKUP MySQL server is stopped, copy the tables (I assume .FRM, .MYI etc?) into the filesystem in /var/lib/mysql/BACKUP_DB (the BACKUP_DB's corresponding directory).
Start the BACKUP MySQL server and ensure the data has loaded correctly using scripts or CLI. When verified, mysqldump the BACKUP_DB database so it can be loaded into the live server:
mysqldump --extended-insert BACKUP_DB > /root/sql/BACKUP_DB.sql
You have now converted your raw backup data into SQL statements which can be loaded into MySQL without downtime (unlike the raw data). Move BACKUP_DB.sql to the live machine.
Import BACKUP_DB.sql into your LIVE MySQL instance as a different database:
mysql BACKUP_DB < /root/sql/BACKUP_DB.sql
You should now have the backup database loaded into MySQL as BACKUP_DB.
Now, dependent on INSERT IGNORE, or REPLACE INTO statements (are you overwriting old data or "filling the blanks" in your indexes?):
mysqldump --no-create-db
--no-create-info --extended-insert --insert-ignore MERGE_SOURCE | mysql BACKUP_DB
Or, for REPLACE INTO action:
mysqldump --no-create-db --no-create-info --extended-insert MERGE_SOURCE | sed 's/INSERT INTO/REPLACE INTO/g' | mysql BACKUP_DB
Alternatively, instead of piping output back into MySQL, send to a file and review the SQL statements.
mysqldump --no-create-db --no-create-info --extended-insert --insert-ignore MERGE_SOURCE > /root/sql/merge.sql
mysqldump --no-create-db --no-create-info --extended-insert MERGE_SOURCE | sed 's/INSERT INTO/REPLACE INTO/g' > /root/sql/merge.sql
Finally, to avoid downtime, dump the first database over the second:
mysqldump BACKUP_DB | mysql CURRENT_DB
You could always lock the database first to prevent data being written to (for example) the z table with a foreign key in the a table (which has already been written over):
FLUSH TABLES WITH READ LOCK;
(perform dump as previous step)
UNLOCK TABLES;
Add the FLUSH command to the start of your dump .sql file and UNLOCK to the end.
Be sure to quadruple check your DB names in this scenario! Ask any follow up question you're unsure of, as this is high risk data mashing stuff. Perform (and note in detail) the exact steps required on a dev server if possible, or virtualize your test, or create a small scale test. Just test. And, of course, take enough backups to cover every data loss eventuality.
Best Answer
You can do this without an intermediary file, as well as resetting the replication pointers in the process, so you don't miss any updates (and have to resync again)
Stop replication the slave
slave> mysql 'slave stop;'
dump the master to the slave, using the
--master-data=1flagmaster> mysqldump -e --master-data=1 --single-transaction $DATABASE | ssh -C user@slave 'mysql $DATABASE'
start replication on the slave
slave> mysql 'slave start'
--master-data=1causes mysqldump to emit theCHANGE MASTER TO ...settings at the top of the dump to set the replication binlog and offset to the exact point in the masters binlog at the time the dump was taken-euses the extended output format, basically multiple sets of value per insert statement, which is more efficent both on the wire, and when being applied to the slave.--single-transationtells mysql to open a transaction over the whole dump, rather than usingLOCK TABLES.