Environment
SUSE Linux Enterprise Server 10
SP1 (i386/x86_64)
Novell Open Enterprise Server 2 (i386/x86_64)
Novell Cluster Services
Novell Open Enterprise Server 2 (i386/x86_64)
Novell Cluster Services
Situation
On a OES2 - 2Node Cluster,
the Cluster Split Brain Detection algorithm does not work as
designed.
In a test scenario where the LAN Cable is pulled from the node having the Master IP address, eventually the voting mechanism needs to determine the node with the proper Link status to survive. The observed behavior on OES2 x86_64 however is that the Master node does not detect the Link Status loss, and as such will not be cast out of the cluster.
For more information on Split Brain Detection, see the TID 10053882 - Gory details of Heartbeats, Split Brains and Poison Pills
In a test scenario where the LAN Cable is pulled from the node having the Master IP address, eventually the voting mechanism needs to determine the node with the proper Link status to survive. The observed behavior on OES2 x86_64 however is that the Master node does not detect the Link Status loss, and as such will not be cast out of the cluster.
For more information on Split Brain Detection, see the TID 10053882 - Gory details of Heartbeats, Split Brains and Poison Pills
Resolution
The fix for this issue has been publicly released in patch oes2-novell-cluster-services-5141.
Additional Information
Internal testing against a number of available network cards from different vendors confirmed the problem to have been resolved in new code, and again working as designed and expected.
Customers running OES environments using 3COMs with the 3c5x9 lan driver are still vulnerable to the problem as the driver for the card is too slow to update it's link status. Considering the mentioned limitation of the mentioned 3c5x9 network drivers in Novell Open Enterprise Server cluster environments alternatives to this specific card need to be considered.
Customers running OES environments using 3COMs with the 3c5x9 lan driver are still vulnerable to the problem as the driver for the card is too slow to update it's link status. Considering the mentioned limitation of the mentioned 3c5x9 network drivers in Novell Open Enterprise Server cluster environments alternatives to this specific card need to be considered.
Change Log
7 Jun 2013 - Removed 'Reported to Engineering' status.