Rollback support in distributed data management systems

A method for providing rollback support for a transaction initiated at a node, the node being one of a plurality of nodes in a distributed system, a reliable replica and a disposable cache copy of a data unit corresponding to the transaction being stored on the plurality of nodes, the data unit comprising a version state and a data state, the method comprising:

• modifying a disposable cache copy of a data unit corresponding to a transaction at a node comprising the disposable cache copy;
• obtaining, in response to the modifying, a modified disposable cache copy at the node;
• propagating a version state and a data state corresponding to the modified disposable cache copy to each node in a plurality of nodes in a distributed system, wherein each node in the of the plurality of nodes comprises at least one reliable replica copy of the data unit, wherein the at least one reliable replica copy is designated as a backup copy for the data unit and is prevented from being used during transaction processing at the node comprising the at least one reliable replica copy, and wherein at least a set of the plurality of nodes each comprise a disposable cache copy of the data unit, wherein the disposable cache copy is a local copy of the data unit specific to the node comprising the disposable cache copy, wherein the disposable cache copy is used during transaction processing at the node and is prevented from being used for backing up the data unit;
• determining, by each of the one or more of the plurality of nodes in response to receiving the version state and the data state, a non-violation, with respect to the transaction, of a set of transaction semantics and a set of replication policies;
• determining that the transaction at the node comprising the modified disposable cache copy was successful;
• committing the transaction to the plurality of nodes in response to the non-violation being determined and the transaction being successful, wherein the committing further comprises:
• revising a version state of the at least one reliable replica copy of the data unit at each of the plurality of nodes based on the version state of the modified disposable cache copy;
• updating a data state of the at least one reliable replica copy of the data unit at each of the plurality of nodes based on the data state of the modified disposable cache copy;
• determining that at least a first node in the plurality of nodes fails to comprise a disposable cache copy of the data unit;
• creating a new disposable cache copy at the first node from the at least one reliable replica copy residing at the first node;
• revising a version state of the new disposable cache copy based on the version state of the modified disposable cache copy;
• updating a data state of the new disposable cache copy based on the data state of the modified disposable cache copy;
• determining that at least a second node in the plurality of nodes fails to comprise a disposable cache copy of the data unit;
• converting the at least one reliable replica copy at the second node into a converted disposable cache copy of the data unit;
• revising a version state of the converted disposable cache copy based on the version state of the modified disposable cache copy;
• and
• updating a data state of the converted disposable cache copy based on the data state of the modified disposable cache copy.

A computer readable storage medium for providing rollback support for a transaction initiated at a node, the node being one of a plurality of nodes in a distributed system, a reliable replica and a disposable cache copy of a data unit corresponding to the transaction being stored on the plurality of nodes, the data unit comprising a version state and a data state, the computer readable storage medium comprising instructions for:

• modifying a disposable cache copy of a data unit corresponding to a transaction at a node comprising the disposable cache copy;
• obtaining, in response to the modifying, a modified disposable cache copy;
• propagating at least one of a version state and a data state corresponding to the modified disposable cache copy to one or more of a plurality of nodes in a distributed system;
• determining, by each of the one or more of the plurality of nodes in response to receiving the at least one of the version state and the data state, a non-violation, with respect to the transaction, of at least one of a set of transaction semantics and a set of replication policies;
• and
• managing a commitment of the transaction at the disposable cache copy based on the non-violation that has been determined, wherein managing the commitment further comprises:
• determining that at least a first node in the plurality of nodes fails to comprise a
• disposable cache copy of the data unit, the first node comprising a reliable replica copy of the data unit is designated as a backup copy for the data unit;
• and
• converting the at least one reliable replica copy at the first node into a converted disposable cache copy of the data unit.