Primary-secondary caching scheme to ensure robust processing transition during migration and/or failover

A method of operating a scoring engine service that comprises a plurality of scoring engine partitions, to maintain scores usable by a behavioral targeting service, the method comprising:

• by each of the plurality of scoring engine partitions, receiving detected event indications indicative of interaction by users with at least one online service, each scoring engine partition being provided events for at least one of a plurality of non-overlapping subsets of the users, and at least one scoring engine partition also being provided events for at least an additional one of said non-overlapping subsets of the users;
• by each of the plurality of scoring engine partitions,
• processing event indications provided to that scoring engine partition to determine, based at least in part thereon, updated scoring data indicative of behavior of the users represented by the events relative to at least one online service and writing the updated scoring data to a persistent scoring engine storage,
• wherein the at least one scoring engine partition being provided event indications for at least an additional one of said particular non-overlapping subsets of the users provides updated scores to the persistent scoring engine storage according to a first writeback caching scheme for updated scores determined from detected events for a first of the particular non-overlapping subsets of the users and according to a second writeback caching scheme for updated scores determined from detected events for the additional one of the particular non-overlapping subsets of the users,
• the method further comprising controlling time-to-live parameters for the first writeback caching scheme independently of time-to-live parameters for the second writeback caching scheme.

A system to maintain scores usable by a behavioral targeting service, comprising:

• an event indication detector that processes event indications, wherein the event indications being processed are indicative of interaction by users generally with at least one online service, wherein some of the event indications are indicative of events usable for generating scoring data for behavioral targeting for providing personalized content, the processing of event indications to detect event indications that are indicative of events usable for generating scoring data for behavioral targeting, and wherein the users include a plurality of particular non-overlapping subsets of users;
• a scoring engine comprising a plurality of scoring engine partitions;
• an event indication provider that provides each of the detected event indications to a separate one of the plurality of scoring engine partitions of the scoring engine, each scoring engine partition being provided detected events for at least one of the particular non-overlapping subsets of the users, and at least one scoring engine partition also being provided detected events for at least an additional one of said particular non-overlapping subsets of the users;
• wherein each of the plurality of scoring engine partitions is configured to:
• process the detected event indications provided to that scoring engine partition to determine, based at least in part thereon, updated scoring data indicative of behavior of the users represented by the detected events relative to the at least one online service and writing the updated scoring data to a persistent scoring engine storage,
• wherein the at least one scoring engine partition being provided detected events for at least an additional one of said particular non-overlapping subsets of the users provides updated scores to the persistent scoring engine storage according to a first writeback caching scheme for updated scores determined from detected events for a first of the particular non-overlapping subsets of the users and according to a second writeback caching scheme for updated scores determined from detected events for the additional one of the particular non-overlapping subsets of the users, and
• control time-to-live parameters for the first writeback caching scheme independently of time-to-live parameters for the second writeback caching scheme.

A computer program product comprising a tangible computer-readable medium having computer program instructions tangibly embodied thereon to configure a plurality of scoring engine partitions of a scoring engine service to maintain scores usable by a behavioral targeting service, the computer program instructions tangibly embodied on the tangible computer-readable medium operable to:

• configure each of the plurality of scoring engine partitions to receive detected event indications indicative of interaction by users with at least one online service, each scoring engine partition being configured to receive events for at least one of a plurality of non-overlapping subsets of the users, and at least one scoring engine partition also being configured to receive events for at least an additional one of said non-overlapping subsets of the users;
• configure each of the plurality of scoring engine partitions, process event indications provided to that scoring engine partition to determine, based at least in part thereon, updated scoring data indicative of behavior of the users represented by the events relative to at least one online service and provide the updated scoring data to a persistent scoring engine storage,
• configure the at least one scoring engine partition configured to receive event indications for at least an additional one of said particular non-overlapping subsets of the users, provide updated scores to the persistent scoring engine storage according to a first writeback caching scheme for updated scores determined from detected events for a first of the particular non-overlapping subsets of the users and according to a second writeback caching scheme for updated scores determined from detected events for the additional one of the particular non-overlapping subsets of the users;
• and
• configure the scoring engine service such that the time-to-live parameters for the first writeback caching scheme are controllable independently of time-to-live parameters for the second writeback caching scheme.