[Linux-cluster] Fencing required for node failover

[Digimer]

On 12/29/2011 04:49 PM, Achint Mehta wrote:
> Hi All, 
> 
> I am using RHCS in RHEL 6.2.
> 
> I am trying to perform a failover for a node in the cluster.
> All the services have fail-over configured on them with recovery method
> set to relocate.
> When the node foes down the services are not relocated to to another nodes.
> 
> Though the node failure is detected by rgmanager:
> ------
> Dec 29 16:20:57 rgmanager State change: pcs_linuxha_1 DOWN
> Dec 29 16:28:25 rgmanager Status Child Max set to 7
> ------
> and fenced has the following logs:
> ------
> Dec 29 16:21:04 fenced fencing node pcs_linuxha_1
> Dec 29 16:21:04 fenced fence pcs_linuxha_1 dev 0.0 agent none result:
> error no method
> Dec 29 16:21:04 fenced fence pcs_linuxha_1 failed
> ------
> 
> 1. Do I require fencing to be enabled to make node failover work
> 2. If yes, what kind of failover device should I add. (all the nodes are
> simple servers.)
> 
> 
> Thanks!
> Achint

Yes, you absolutely needs fencing.

As soon as a node is lost, fenced informs dlm which then stops providing
locks. Only when the fence succeeds is dlm informed and will again issue
locks. In turn, rgmanager uses dlm, so with dlm not providing locks,
rgmanager can't recover services.

See this for a more specific explanation;

https://alteeve.com/w/2-Node_Red_Hat_KVM_Cluster_Tutorial#Concept.3B_Fencing

-- 
Digimer
E-Mail:              digimer at alteeve.com
Freenode handle:     digimer
Papers and Projects: http://alteeve.com
Node Assassin:       http://nodeassassin.org
"omg my singularity battery is dead again.
stupid hawking radiation." - epitron