Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I am preparing for setting up a MySQL master-slave or master-master relationship. Right now I have a single MySQL production server, and of course I don't want a lot of downtime while I connect the slave.

Is there no way I can make connect an empty slave and let it "slowly" synchronize data from the master, until they are identical?

I have noticed that I can take a transactional dump with mysqldump on the master and then import that into the slave, but by the time the slave has imported the dump, a lot of new rows will have been written and how will the slave get these?

I hope I am missing something obvious here, but extensive googling gives such advice such as "since this will result in less downtime in the future, some downtime now maybe isn't such a bad thing". But I would really like to avoid that.

Related question for MyISAM, but I use InnoDB.

share|improve this question
add comment

2 Answers

up vote 7 down vote accepted

If you're using 100% InnoDB then you're in luck. You can use XtraBackup to make a full backup of your master database without any downtime or any table locking. This will be a consistent snapshot-style backup, the same as the sort that you get when you do the FLUSH TABLES WITH READ LOCK or --master-data options.

The XtraBackup tool also drops an extra file in the backup directory that contains the MASTER_LOG_POS and MASTER_LOG_FILE information you need to start replication on the slave.

Once you're done backing up, you will need to run XtraBackup's --prepare option on the backup, load it into the slave, start the slave MySQL process back up and tell it the new MASTER_LOG_POS and MASTER_LOG_FILE values it needs.

You will want skip-slave-start in your my.cnf before you start the slave up.

Also bear in mind that the mysql schema is MyISAM by default (and if memory serves correctly it can only be MyISAM) so you will still have to be careful not to make any changes to any of those tables while running the backup. As long as you stick to that rule, the master information will still be correct.

It's often a good idea to ignore the mysql schema in your my.cnf on the slave and only ever create users with SELECT privileges. Inconsistent and out-of-sync slaves are hard to detect and a pain to deal with, even when using the tools that Percona (and Maatkit before them) provide for this.

Edit:

Although you said you are using InnoDB, for completeness there is another way if you are using MyISAM tables. If you have a volume manager with snapshotting (such as ZFS or LVM), you can run a FLUSH TABLES WITH READ LOCK followed by a SHOW MASTER STATUS, create a snapshot and run UNLOCK TABLES. The downtime should be fairly minimal. For comparison, the cron job last night that did this to backup one of our databases took 6 seconds to create the snapshot which is the bit where the database is "down" and 27 minutes to copy the files from the snapshot to the backup server.

share|improve this answer
    
XtraBackup seems to work pretty well with MyIsam too, albeit a little slower. –  Prof. Falken Nov 21 '11 at 9:12
add comment

It is not possible to start a mysql replication "from scratch".

Instead, you could use mysqldump with the --master-data=2 parameter when dumping your databases - i.e. like this:

mysqldump --master-data=2 --all-databases --opt -p > myinitialdump.sql

This command will lock all tables within the affected databases during the dump and write the replication coordinates into the dump's header in a commented-out fashion like this:

-- CHANGE MASTER TO MASTER_LOG_FILE='mysql-bin.000018', MASTER_LOG_POS=106;

After your import is complete, you can manually run the "change master to" command spiced with the host name of your master and the authentication data - the replication will lift off at the point of the dump.

Please keep in mind that the mysqldump process will incur "downtime" in itself due to locking contention - it places read locks on all tables (similar to what a FLUSH TABLES WITH READ LOCK command would do) for the entire duration of the dump. Thus, no requests writing to any of the tables will return unless the dump is complete. Also, the write requests probably also would block any subsequend read requests unless you've specified low_priority_updates in your MySQL configuration or have issued SET GLOBAL LOW_PRIORITY_UPDATES=1 before the dump.

share|improve this answer
1  
Actually, you can start "from scratch" as long as you have had binary logging turned on from the very start and you have never deleted a binary log. This is pretty unlikely but I have actually seen it done. –  Ladadadada Nov 18 '11 at 12:58
    
@Ladadadada, I figured, but if we were that good we would have actually made a server-slave setup from the beginning. :-) –  Prof. Falken Nov 18 '11 at 13:31
    
It's not good - it is on the edge to insane. –  the-wabbit Nov 18 '11 at 13:49
    
@syneticon-dj ha, why? At this point I almost wish I was insane. –  Prof. Falken Nov 18 '11 at 13:51
    
because for any significant age and update rate of a database, the binary logs will outgrow the database's data by magnitudes. –  the-wabbit Nov 18 '11 at 19:07
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.