At the moment, if a HA-pair is turned off, all data from the ha-pair will be un-accessible. Only the data from the surviving nodes will be served.
Looking at the release notes for 8.3.1: *If epsilon is held by a node in a compromised HA pair, it is reassigned automatically to a node in a noncompromised HA pair. An HA pair is considered to be compromised if either node goes offline.*
--tmac
*Tim McCarthy* *Principal Consultant*
On Thu, Dec 3, 2015 at 4:45 PM, Peter Choi shpeterchoi@gmail.com wrote:
Yes manually moving epsilon is feasible in a planned downtime scenario. What about in a disaster situation - if a pair goes down unexpectedly, will epsilon me auto migrated to either of the surviving nodes?
On Dec 3, 2015, at 4:39 PM, Gelb, Scott scott@red8.com wrote:
I tested on 8.3.1 and when an ha-pair is degraded, epsilon moves to the other ha-pair when one of the nodes goes down, however I would move epsilon manually to a node in the other ha-pair for this type of maintenance and not assume it will move. Then you have 2 nodes plus epsilon and continue to serve data for the aggregates/volumes on those nodes…the data on the other ha-pair shutdown will be inaccessible and the cluster will not be in a healthy state.
*From:* toasters-bounces@teaparty.net [ mailto:toasters-bounces@teaparty.net toasters-bounces@teaparty.net] *On Behalf Of *Peter Choi *Sent:* Thursday, December 3, 2015 1:35 PM *To:* toasters@teaparty.net *Subject:* Shutting down a controller pair in a 4-node cluster.
Hi,
Please consider the following scenario, as we are conducting testing. I've done my research and received varying information from support , so I thought to get this audience's take on this.
Basically we have a flexpod.
2 cabinets;
Each cabinet consists of: CN1610, Nexus, FI, 2 x FAS8020 (cDOT 8.3.1P1),
UCS blades;
Controllers are joined together in a 4-node cluster configuration;
Cabled in such a manner that we have controller HBA and switch port
redundancy.
If we were to shutdown a controller pair (2 nodes that are HA partners to each other), effectively shutting down 1 of the 2 cabinets.
- Will the data on the disks attached to the other two controller nodes
(the surviving HA pair) still be served and accessible?
- What happens if epsilon resides on one of the nodes that are shut down.
Will it be automatically shifted to a surviving member in the cluster?
Thanks.
Toasters mailing list Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters