[DRBD-user] Re quest comments: DRBD to cold BackUp DMS with minimal DownTime
bart.coninckx at telenet.be
Fri Nov 12 11:51:52 CET 2010
On Thursday 11 November 2010 13:07:09 I. Schumacher wrote:
> I am administrator of an Alfresco DMS running on an dedicated ubuntu server
> and is backedup to a streamer via a simple cron job.
> Unfortunately my predecessor (who set up this backup process) did not know
> that streaming a backup while the DMS is running causes mysql database
> inconsistencies preventing the DMS to boot after recovery. boo-hoo
> Running a proper cold back requires approx 2h while the DMS is shutdown.
> Thats just to long.
> One way would be to set up a slave-read-only-dms-site with mysql
> replication which can be shutdown for backup anytime while the master
> system is still running. In my opinion this is a waste of ressources
> because the slave site got to be up all the time, only building indices
> out of the replicated mysql data.
> In the course of an Upgrade to a newer version of the DMS I plan to
> implement a new backup process using DRBD involving a well equipped
> Master-DMS-Site and a MiniDRBD-Site for backup on another machine.
> 1. I have two virtual machines on two separate physical machines. VM1=
> MasterSite VM2=MiniDRBD
> 2. VM1 is running Ubuntu Server, RDBD, iptables, sshd, mysqld,
> oOffice-headless and tomcat.
> 3. Residents of VM2 are Ubuntu Server, RDBD, sshd and iptables
> 4. VM1 and VM2 use a virtual HardDisk /dev/sdb each residing on the
> corresponding physical machine
> 5. RDBD repilicates "VM1/dev/sdb0" to "VM2/dev/sdb0" via /dev/drdb0
> 6. VM1 stores all DMS relevant data including mysql database and indices on
> To stream a cold BackUp with minimal DownTime I intend to use the following
> 1. crontab a shell script on VM1 which
> => stops DMS closing the database properly
> => RDBD replicates the closed state of the database to VM2/dev/sdb0 via
> => stops replication via "drbdadm disconnect MyFS"
> => starts the DMS again
> => waits for 2.5h and starts replication via "drbdadm connect MyFS"
> => exits
> 2. Five Minutes after 1. started I crontab a shell script on VM2 which
> => mounts VM2/dev/drdb0 to e.g. /media/drdb0
> => streams all files to tape via "tar cvf - /media/drdb0/* | dd
> of=/dev/str0/n block-size=4096"
> => umounts /media/drdb0
> => exits
> (I think that:)
> Streaming a BackUp that way reduces the downtime of the Master
> (VM1/alfresco) to about 10 seconds instead of 2 hours. Additionally it (i
> hope) nearly halves required CPU time and RAM usage in comparison to the
> mysql-replication approach that is recommended by the Alfresco-DMS vendor.
> (:but I may be wrong)
> Some general questions:
> Q1. I wonder if DRDB will autosync all changes made on the master during
> the "disconnect" to the slave after a "drbdadm connect MyFS" (in script 1.
> on VM1).
> Q2. How long will it take drbd to replicate the closed state of a 150mb
> database to VM2 so I can stop replication? (1GB/s LAN)
> Q3. What do you think would be the minimum specs for my miniRDBD machine
> (VM2). (128RAM,1core,256swap?)
> Q4. Do you have any suggestions how to use drbd a smarter way for my backup
> Q5. <Non drbd question> What is the best way to invoke all commands from
> only one script on VM1? Commands on VM2 will have to be executed as root.
> How can I do that without storing my root password as plaintext inside the
> script. I know I did that some time ago but I forgot how...
> I hope you can help me.
> Thanks a lot in advance. ;)
Would using LVM snapshotting be a possibility? If you use it for your MYSQL
storage, it can provide you with a stable readonly copy when you take it after
shutting down MYSQL. If you use it for your VM storage, you could copy your
entire guest after pauzing or saving it.
More information about the drbd-user