5

I have following gluster volume, details are as follows

Volume Name: geo-vol
Type: Distribute
Status: Started
Number of Bricks: 1
Transport-type: tcp
Bricks:
Brick1: bst:/backup
Options Reconfigured:
geo-replication.indexing: on

I am mounting this volume on the same machine as nfs mount and brick1 is also on the same machine and than using geo-replication to mirror it to the backup server.

As you can see from my setup I am using glusterfs for almost realtime backup.

I just need reliable way to backup my data to secondary server, earlier I was using rsync but it started taking a lot of memory as number files increased, so we switched to gluster, when we tried real time replication it was hampering performance of the server, so in the last resort we went with ge-replication, one issue which we are facing right now is the cpu consumption of gluster is way high, I asked this question to gluster mailing list but no update.

vishy
  • 61
  • 1
  • 3
  • I'm not going to post an answer because I don't know that much about gluster, but I must point out that you are making a mistake made my many a novice sysadmin, which is to equate redundancy with a backup. What you have is redundancy, your data is updated in multiple locations in almost real time. This also means that erroneous modifications or accidental deletions are similarly updated in all locations, almost certainly before you even realise there's an issue. Do yourself a massive favour and read this before you regret it: http:// serverfault.com/questions/2888/why-is-raid-not-a-backup – ThatGraemeGuy Aug 08 '12 at 11:42
  • Thanks for the link, I have daily and weekly backup through snapshots, what I want to achieve here is for the failure through the day for which I said I was using rsync but than due to issues I moved to glusterfs and if you read it right I am using geo-replication which is not realtime, I am not using raid1 configuration in gluster. – vishy Aug 08 '12 at 11:59
  • Cool. I don't understand the finer implementation details of gluster, my point was more that you are using the term "backup" to refer to a configuration that is only providing redundancy, and it happens far too often that folks assume they only need redundancy. – ThatGraemeGuy Aug 08 '12 at 14:30

1 Answers1

3

As per http://community.gluster.org/q/running-client-server-on-the-same-set-of-nodes/ :

It's fairly common to run client and server processes on the same set of nodes. In fact, GlusterFS servers will mount a volume as a client to perform certain operations. In general this works well, but it is possible to run into various forms of contention between the server and client processes.

S19N
  • 1,693
  • 1
  • 17
  • 28