as said:
use a replication slave dedicated for backups
you can even let a slave write a binlog and
sync another slave with this one
* rsync backups working with diff
* they are extremly fast after the first time
* a dedicated backup-slave has ZERO impact
i am doing rsync-backups of 1.5 TB data over a WAN
link since years each day and the real traffic is
between 2 and 5 GB each day
Am 01.11.2012 16:53, schrieb machiel.richards@stripped:
> Well, the biggest problem we have to answer for the clients is the following:
> 1. Backup method that doesn't take long and don't impact system
> 2. Restore needs to be done on a quick as possible way in order to minimize
> downtime.
>
> The one client is running master - master replication with master server in usa, and
> slave in south africa. They need master backup to be done in the states.
>
> Sent via my BlackBerry from Vodacom - let your email find you!
>
> -----Original Message-----
> From: Reindl Harald <h.reindl@stripped>
> Date: Thu, 01 Nov 2012 16:49:45
> To: mysql@stripped<mysql@stripped>
> Subject: Re: Mysql backup for large databases
>
> good luck
>
> i would call snapshots on a running system much more dumb
> than "innodb_flush_log_at_trx_commit = 2" on systems with
> 100% stable power instead waste IOPS on shared storages
>
> Am 01.11.2012 16:45, schrieb Singer Wang:
>> Assuming you're not doing dumb stuff like innodb_flush_log_at_tx=0 or 2 and etc,
> you should be fine. We have been
>> using the trio: flush tables with read lock, xfs_freeze, snapshot for months now
> without any issues. And we test
>> the backups (we load the backup into a staging once a day, and dev once a week)
>>
>> On Thu, Nov 1, 2012 at 11:41 AM, Reindl Harald <h.reindl@stripped
> <mailto:h.reindl@stripped>> wrote:
>>
>> > Why do you need downtime?
>>
>> because mysqld has many buffers in memory and there
>> is no atomic "flush buffers in daemon and freeze backend FS"
>>
>> short ago there was a guy on this list which had to realize
>> this the hard way with a corrupt slave taken from a snapshot
>>
>> that's why i would ALWAYS do master/slave what means ONE time
>> down (rsync; stop master; rsync; start master) for a small
>> timewindow and after that you can stop the slave, take a
>> 100% consistent backup of it's whole datadir and start
>> the slave again which will do all transactions from the
>> binarylog happened in the meantime
Attachment: [application/pgp-signature] OpenPGP digital signature signature.asc
