Gluster users - Jun 2011 - Crossover cable: single point of failure?

Dear community,

 I have a 2-node gluster cluster with one replicated volume shared to a 
 client via NFS. If the replication link (Ethernet crossover cable) 
 between the Gluster nodes breaks, I discovered that my whole storage is 
 not available anymore.

 I am using Pacemaker/corosync with two virtual IPs (service IPs exposed 
 to the clients), so each node has its corresponding virtual IP, and if 
 one node fails, corosync assigns the failing IP to the other running 
 node). This mechanism works pretty good so far.

 So, I have:

   gluster1: IP 10.196.150.251 and virtual IP 10.196.150.250
   gluster2: IP 10.196.150.252 and virtual IP 10.196.150.254

 Now I am using DNS round-robin to distribute the load on both gluster 
 nodes (name: gluster.mycompany.tld).

 If one node goes down, the virtual IP is handed over to the remaining 
 node, and the client works without any disruption. However, if the 
 replication link between the gluster nodes breaks, we examined a service 
 disruption. The client was then unable to write data to the cluster. The 
 replication links between gluster nodes seems to be a single point of 
 failure. Is that correct?

 Daniel

Can you please share NFS and brick logs from the duration of the link going
down? Gluster should have worked in the situation you described.

Avati

On Fri, Jun 10, 2011 at 3:27 PM, Daniel Manser <daniel at clienta.ch>
wrote:
> Dear community,
>
> I have a 2-node gluster cluster with one replicated volume shared to a
> client via NFS. If the replication link (Ethernet crossover cable) between
> the Gluster nodes breaks, I discovered that my whole storage is not
> available anymore.
>
> I am using Pacemaker/corosync with two virtual IPs (service IPs exposed to
> the clients), so each node has its corresponding virtual IP, and if one
node
> fails, corosync assigns the failing IP to the other running node). This
> mechanism works pretty good so far.
>
> So, I have:
>
>  gluster1: IP 10.196.150.251 and virtual IP 10.196.150.250
>  gluster2: IP 10.196.150.252 and virtual IP 10.196.150.254
>
> Now I am using DNS round-robin to distribute the load on both gluster nodes
> (name: gluster.mycompany.tld).
>
> If one node goes down, the virtual IP is handed over to the remaining node,
> and the client works without any disruption. However, if the replication
> link between the gluster nodes breaks, we examined a service disruption.
The
> client was then unable to write data to the cluster. The replication links
> between gluster nodes seems to be a single point of failure. Is that
> correct?
>
> Daniel
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20110610/db4bb551/attachment.html>