trigger for a whole lost network MultiNiCB
In a common clusterenvironment there are sometimes failure situations where is no need to switch the service groups from one node to a another node in the cluster. one of this situations is a whole failed network. If the network fails simultaneously over all nodes the cluster framework attempt to start the Servicegroups in one other node in cluster. This behaviour drive the cluster to unnecessary actions and the servicegroup with the application becoming faulted on all clusternodes. In comparision to a single server this is more worst behaviour.
Therefore there is a need for a special trigger which freezes all servicegroups when the MultiNICB ressource fails simultaneously on all cluster nodes . This would prevent a lot of work after a whole lost network for all customers particularly if the cluster configured with a lot of servicegroups.
Comments 3 Comments • Jump to latest comment
Please let me know if I have understood the idea as intended:
"When there is a failed network which impacts all the nodes in the cluster, there is no benefit in attempting a failover since the standby node is impacted by the same issue. Hence, VCS should have the ability to detect that the whole network is lost and therefore should freeze the SGs rather than attempting a failover"
Some of the resilience characteristics of VCS, which isolate the nodes from one another, might restrict the ability of one node to be aware of network failure across all nodes of the clusters. However, we will definitely consider the idea for further review to address the failover-in-vain issue.
With "a whole failed network" we mean the public network, the heartbeat network is still working.
From the several votes in agreement, the demand for this ability is evident. From an internal discussion, it is clear that we would need involved changes to enable this functionality - we will consider this in our roadmap for our next release after 6.0. Thanks for the helpful suggestion!
Would you like to reply?
Login or Register to post your comment.