Suicide among well-mannered cluster nodes experiencing heartbeat failure

摘要

Methods for re-configuring a cluster computer system of multiple or more nodes when the cluster experiences communications failure. First and second nodes of the cluster have respective channel controllers. A SCSI channel and the controllers communicatively connect the multiple nodes. When a node becomes aware of a possible communications failure, the node attempts to determine the authenticity the failure and responds according to the determined authenticity.;According to one method, a first node detects heartbeat node-to-node communications failure on the channel and then tests a physical drive on the channel. If the testing is successful, the node kills the other node. If the testing is unsuccessful, the first node commits suicide.;In one embodiment, the coupling includes multiple channels communicatively coupling the first and second nodes and the first node selecting one of the channels for node-to-node communications. In this environment, choosing a physical drive involves testing node-to-node communications on another of the channels if no physical drive is online on the channel (and terminating the re-configuring method). If a drive is available, the first node uses the first physical drive online on the channel for testing.;In another method, the second node initially detects communications failure and communicates that by attempting to negotiate wih the first node for a new configuration of the computer system. The first node tests a physical drive in response and negotiates with the second node if the testing was successful. If the testing was unsuccessful, the first node commits suicide.