[Gluster-users] Replicated Volume (mirror) on 17 nodes.
ravishankar at redhat.com
Thu Feb 25 08:20:24 UTC 2016
On 02/25/2016 11:42 AM, Simone Taliercio wrote:
> Hi Ravi,
> Thanks a lot for your prompt reply!
> 2016-02-25 6:07 GMT+01:00 Ravishankar N <ravishankar at redhat.com>:
>> I don't know what your use case is but I don't think you want to create so many replicas.
> I need to scale my application on multiple nodes because we have to
> tackle an high number of requests per second. The application is
> hosted on AWS EC2 instances and each one uses EBS.
> Each instance needs to read some local files. That's why I'm currently
> replicating all the files on each instance. So far I had max 3 nodes.
>> Why not just create a glusterfs volume of replica-2 or replica-3 or even an arbiter volume, and mount them on all 17 nodes?
> Here I'm definitely missing some basics (sorry for that): what are the
> steps to set-up 3-replicas (+1 arbiter) but still allow all 17 nodes
> to retrieve the files ?
The steps are almost the same as replica -3, except for a small change
in the syntax: `gluster volume create <volname> replica 3 arbiter 1
<host1:brick> <host2:brick> <host1:brick> <host3:brick>`
> Are they transferred on demands over the
> network ?
It's just like accessing an NFS share. You mount the volume on any
machine and you can perform I/O.
> So far I followed those steps:
> - mount an empty "hard disk" to the instance
> - format it in xfs
> - create directory for one brick
> - add peers
> - create 1 volume with "... replica 3 <host1:brick> <host2:brick>...."
> - mount the volume on a specific path
> And then I can see one file that is created on one machine being
> replicated on all the others. I have a limited vision at this use case
Right. You can mount the replica 3 volume that you just created on any
node. Like I said it's just like accessing a remote share. Except that
the 'share' is a glusterfs volume that you just created.
If I understand your use case correctly, you would need to create a
glusterfs volume based on 3 EC2 instances and then mount the volume on
all of the (17?) instances on which your application runs.
>> I think replica 4 is what some people in the community have used at the max but even that is an overkill IMO.
> Help :)
More information about the Gluster-users