RE-FORMING AN APPLICATION CONTROL TREE WITHOUT TERMINATING THE APPLICATION

摘要

A reconnection system re-forms a control tree for an application that is executed in parallel without terminating execution of the application. The reconnection system detects when a node of a control tree has failed and directs the nodes that have not failed to reconnect to effect the re-forming of the control tree without the failed node and without terminating the application. Upon being directed to reconnect, a node identifies new child nodes that are to be its child nodes in the re-formed control tree. The node maintains the existing connection with each of its current child nodes that is also a new child node, terminates the existing connection with each of its current child nodes that is not also a new child node, establishes a new connection with any new child node that is not a current child node, and directs each new child node to reconnect.

主权项

1. A method performed by a node in a computing system for responding to a failure during parallel execution of instances of an application at multiple nodes without terminating the application, the nodes and connections between the nodes forming a control tree in which parent nodes are connected to child nodes, the method comprising:
receiving at a parent node an indication to reconnect the control tree with new child nodes as child nodes of the parent node; for each current child node of the parent node that is not a new child node, terminating the connection between the parent node and the child node; and for each new child node,
when the new child node is a current child node of the parent node, maintaining the connection with the new child node; andwhen the new child node is not a current child node of the parent node, establishing a connection with the new child node so that the new child node is a child node of the parent node.