Am 02.03.2012 17:08, schrieb mail...@securitylabs.it: >> how do you do your backups? > Bacula with Accurate backup
another software needed >> i stop the slave per cron, make the backup with rsync >> and start the slave again having a 100% consistent >> copy > You need two server or at least two MySQL instance only to obtain a backup no problem on a VMware-HA-Cluster >>> You don't have to backup a big database, just many small files >> and you can make a 100% constistent backup pf this >> many small files while the service is running? > Sure, Maildir does not lock files to work. your backup is not anatomic there are changes between start/end a halted slave is >> so you can restore easy and capillar, how you restore one message or one >> mailbox with dbmail? >> simply connect to the backup-machine running a second instance >> parallel to the replication-slave and drag whatever message >> whereever i like > The slave is replicated, so if you have to restore a mailbox deleted on the > master the data has > already been deleted on the slave (otherwise it's not a replication), you > have to > restore the entire db from a previous backup you did not read the "second instance" i was smart enough to setup at least two runnings mysqld on the backup-machine, one is a readonly slave and the other one is a daily clone of this running dbmail-imapd additionally there is a "last-week" backup where i can switch with one single command the imapd > For sure the big advantage is that replication is easy and you can make an > active/passive system so that if the active note crashes you can be online > in a few seconds without using clustered file system, network file system and > so on. this is another point but does not matter on a HA-Cluster if the host crashs the active instance is started on another one and a notify fired up to the admin as said in days where you can have > 20 servers on two hosts with a dual-controller SAN a backup-instance with reduced ressource configuration is no problem and the benfit of this setup is that you can test software-upgrades with live data instead generic test failing in the real-life, simply update the backup-vm and run as much tests as you like on a snapshot roll back the snapshot -> the slave tales only a short-time to be synchron again
signature.asc
Description: OpenPGP digital signature
