Because the Cluster Edition coordinates its membership changes with VCS and relies on the SF for Sybase CE to perform I/O fencing, the Cluster Edition membership manager waits for VCS to reconfigure its membership and perform I/O fencing. However, during some failures, Adaptive Server requires I/O fencing even if the situation does not result in a VCS membership reconfiguration. To acquire the I/O fencing, Adaptive Server passes a message through the quorum device to the VCS agent on the node to be fenced. When it receives the message, the VCS agent panics the node, which triggers a VCS reconfiguration.
How the VCS system reacts during a failure depends on the situation:
Host node crashes – detected by VCS. VCS reconfigures cluster membership and performs I/O fencing from the failed node. The Cluster Edition receives the new membership from VCS and performs recovery.
Host node hangs – results in a VCS LLT timeout and a Cluster Edition heartbeat timeout. VCS reconfigures and performs I/O fencing. The Cluster Edition receives the new membership from VCS and performs recovery. The node that is hanging cannot write to the shared storage and eventually panics.
Instance fails or hangs – the remaining instances detect a heartbeat failure and perform membership arbitration. The cluster passes a message to the VCS agent on the host node on which the instance failed, which triggers a host node panic. From this point, the VCS acts in the manner described in the “Host node crashes” situation described above.
Instance shutdown with nowait – the Cluster Edition does not trigger a VCS membership change in VCS membership mode.
Loss of storage access – the Cluster Edition is unaffected if you configured multipathing and at least one path to storage remains viable. However, if all paths to storage are lost, VCS induces a host panic, which the cluster handles in the same manner described in “Host node crashes” above.
Loss of interconnect – response depends on the availability of an alternate communication channel. Both Adaptive Server and VCS can switch traffic to an alternate channel. If Adaptive Server has no viable channels and VCS does, the cluster reacts in the manner described in “Instance fails or hangs,” above. If VCS has no viable channels, it acts in the manner described in the “Host node crashes” situation described above
