Method for ensuring operation during node failures and network partitions in a clustered message passing server

For a message system for delivering data in the form of messages between message clients, the message system comprising a server cluster containing a group of message manager nodes with means for storing and distributing messages a method for ensuring operation during node failures and network partitioning, the method comprising:

• Maintaining, in each running message manager node, a cluster state data set comprising information about the nodes present in the cluster
• Repeatedly evaluating, in each running one of said message manager nodes, a view state data set comprising information about the server nodes able to communicate with said message manager node,
• Comparing said cluster state data set with said view state data set and evaluating if all message managers are able to communicate with said message manager node,
• If not all message managers are able to communicate with said message manager node, determining an operational state for publish/subscribe style messaging out of at least two different states, where
• a first one of said at least two different states is an non-retaining state, the message managers being in said non-retaining state being allowed to delete publish/subscribe messages as soon as it has determined that all eligible subscribes have received copies of those messages, and
• a second one of said two different states is a retaining state, the message managers being in said retaining state not being allowed delete publish/subscribe messages prior to the expiry of the individual messages or the message manager changes to the non-retaining state and determines that all eligible subscribers have received copies of those messages,
• and where the operational state is determined in a manner that of any two message manager nodes that are responsible for dispatching the same publish/subscribe messages not being able to communicate with each other, both must be in a retaining state.

For a message system for delivering data in the form of messages between message clients, the message system comprising a server cluster divided into sub-clusters containing a group of message manager nodes with means for storing and distributing messages a method for ensuring operation during node failures and network partitioning, the method comprising:

• Maintaining, in each running message manager node a sub-cluster state data set comprising information about the servers present in the cluster
• Repeatedly evaluating, in each running one of said message manager nodes, a view state data set comprising information about the server nodes able to communicate with said message manager node,
• Comparing said sub-cluster state data set with said view state data set and evaluating if all message managers of the sub-cluster are able to communicate with said message manager node,
• If not all message managers of the sub-cluster are able to communicate with said message manager node, determining a point-to-point operation state out of at least two different states, where
• a first one of said at least two different states is an unrestricted state, the message servers being in said unrestricted state being allowed to dispatch point-to-point messages sent prior to and since the last change of the view state of the current network, and
• a second one of said two different states is a restricted state, the message servers being in said restricted state not being allowed to dispatch point-to-point messages sent prior to the last change of the view state of the current network,
• and where the operational state is determined in a manner that of two message manager nodes of the same sub-cluster not being able to communicate with each other, at most one can be in an unrestricted state.

For a message system for delivering data in the form of messages between message clients, the message system comprising a server cluster containing a group of message manager nodes with means for storing and distributing messages a method for guaranteeing JMS semantics during node failures and network partitioning, the method comprising:

• Maintaining, in each running message manager node, a cluster state data set comprising information about the nodes present in the cluster
• Repeatedly evaluating, in each running one of said message manager nodes, a view state data set comprising information about the server nodes able to communicate with said message manager node,
• Comparing said cluster state data set with said view state data set and evaluating if all message managers are able to communicate with said message manager node,
• If not all message managers are able to communicate with said message manager node, determining an operational state for point-to-point style messaging out of at least two different states, where
• a first one of said at least two different states is an unrestricted state for normal operation, the message managers being in said unrestricted state being allowed to dispatch point-to-point messages sent both prior to and since the last change of the view state of the current network, and
• a second one of said two different states is a restricted state, the message managers being in said restricted state not being allowed to dispatch point-to-point messages sent prior to the last change of the view state of the current network,
• and where the operational state is determined in a manner that of any two message manager nodes that are responsible for dispatching the same point-to-point messages not being able to communicate with each other, at most one can be in an unrestricted operational state,
• and further comprising, if not all message managers are able to communicate with said message manager node, determining an operational state for publish/subscribe style messaging out of at least two different states, where
• a first one of said at least two different states is an non-retaining state for normal operation, the message managers being in said non-retaining state being allowed to delete publish/subscribe messages as soon as it has determined that all eligible subscribes have received copies of those messages, and
• a second one of said two different states is a retaining state, the message managers being in said retaining state not being allowed to delete publish/subscribe messages prior to the expiry of the individual messages or until the message manager changes to the non-retaining state and determines that all eligible subscribers have received copies of those messages,
• and where the operational state is determined in a manner that of any two message manager nodes that are responsible for dispatching the same publish/subscribe messages not being able to communicate with each other, both must be in a retaining state.

A message system for delivering data in the form of messages between message clients, comprising a server cluster containing a group of message manager nodes with means for storing and distributing messages a method for ensuring operation during node failures and network partitioning, wherein each message manager node comprises means for

• Maintaining, in each running message manager node, a cluster state data set comprising information about the nodes present in the cluster
• Repeatedly evaluating, in each running one of said message manager nodes, a view state data set comprising information about the server nodes able to communicate with said message manager node,
• Comparing said cluster state data set with said view state data set and evaluating if all message managers are able to communicate with said message manager node,
• If not all message managers are able to communicate with said message manager node, determining an operational state for point-to-point style messaging out of at least two different states, where
• a first one of said at least two different states is an unrestricted state for normal operation, the message managers being in said unrestricted state being allowed to dispatch point-to-point messages sent both prior to and since the last change of the view state of the current network, and
• a second one of said two different states is a restricted state, the message managers being in said restricted state not being allowed to dispatch point-to-point messages sent prior to the last change of the view state of the current network,
• and wherein said means for determining an operational state out of at least two states are configured in a manner that of two message manager nodes that are responsible for dispatching the same point-to-point messages not being able to communicate with each other, at most one can be in an unrestricted operational state.

A computer program product comprising a computer usable medium having computer readable program code means embodied therein for enabling a computer to serve as a message manager in a server cluster, the program product comprising computer readable code means for enabling the computer

• To maintain, in each running message manager node, a cluster state data set comprising information about the nodes present in the cluster
• To Repeatedly evaluate, in each running one of said message manager nodes, a view state data set comprising information about the server nodes able to communicate with said message manager node,
• To compare said cluster state data set with said view state data set and to evaluate if all message managers are able to communicate with said message manager node,
• If not all message managers are able to communicate with said message manager node, to determine an operational state for point-to-point style messaging out of at least two different states, where
• a first one of said at least two different states is an unrestricted state for normal operation, the message managers being in said unrestricted state being allowed to dispatch point-to-point messages sent both prior to and since the last change of the view state of the current network, and
• a second one of said two different states is a restricted state, the message managers being in said restricted state not being allowed to dispatch point-to-point messages sent prior to the last change of the view state of the current network,
• and where the means for determining an operational state are programmed in a manner that of any two message manager nodes that are responsible for dispatching the same point-to-point messages not being able to communicate with each other, at most one can be in an unrestricted operational state.