CentOS And GlusterFS

Home » CentOS » CentOS And GlusterFS
CentOS 3 Comments

Hi list, I’m planning to setup a two-node “cluster” with file replication and samba. I’ve used drbd in the past with success, but now there isn’t drbd on the official repo but only from elrepo. I want follow CentOS Line and than switch to GlusterFS.

Said that, I’m new to GlusterFS.

I have two hosts with a 2tb disk space for replica. I’ve ridden that glusterfs is not compatible with smb/cifs/nfs and that for this purpose I must mount locally the volume and share it with samba and the hosts must act as client and server at the same time.

It is possibile shares a mounted disk on /data with samba, and apply a replica volume on /data without using glusterfs export?

Example:
[Host1] [Host2]
/data -> smb /data -> smb gluster_replica<---------------->gluster_replica

Ip and services will be managed from corosync+pacemaker.

There will be problems and complication with this configuration about replication/sync?

Thanks in advance.

Alessandro.

3 thoughts on - CentOS And GlusterFS

  • I can’t speak to CentOS 6 but I have a few C5 HA clusters that use drbd for most cluster resources but use glusterfs for /home (and thus ~/public_html). Works fine.

    I can’t check right now but I think I was using the native gluster client rather than NFS.

    Going by memory, I think the drbd main repo was supposed to be fine for use with CentOS as well. If you’re replacing a system that worked well with drbd, I’d say stick with it. Glusterfs is fine, but drbd is less cantankerous and has fewer failure modes.

    I’m just waiting for CentOS 7 to come out and then those C5 clusters will get upgraded straight to C7.

    Devin (who hates typing with his thumbs on phone keyboards …)

  • A caution–a two-node replicated cluster is a poor choice. If something gets into a split-brain condition (easier to happen than it should be)
    there are only two different nodes, and so there can be no voting. Using a three-node replicated cluster is the minimum that can be recommended at this point. (You don’t want to get into a split-brain condition, because it will give you a headache that makes it feel like your brain is splitting.)

    Gluster is beginning work on a different way of writing the replicated data, that will be much less likely to result in split-brain,

    I’m not sure what you’re trying to do here, but nfs sharing is built into gluster, and turned on by default. What Operating System will be using the gluster volume?

    Be sure to map the gluster file system to /data, NOT the underlying bricks.
    Mapping the bricks is a common newbie mistake. Gluster does NOT
    replicate changes to the individual bricks. Only changes to the mounted glusterfs volume itself are replicated reliably. Be sure that when you type “mount”, the line that describes /data looks like:
    localhost:data /data glusterfs It HAS TO say glusterfs in order to be shared.

    There will be a separate line in the output of “mount” that will say something like

    /dev/sda2 /brick/data xfs

    That is the brick, and your samba share must NOT point to /brick/data.

    Ted Miller