Lendingclub Corporation

Lendingclub Bank, National Association

Abhijit Karpe

Paul Strack

Srinivasa Ambikapathi

Yana Nikitina

Hyunsuk Han

Raul Acevedo

David Taubler

Techniques are described herein to handle situations in which multiple systems can change different copies of the same data item. Optimistic locking and time stamps are used to ensure consistency between the systems without incurring the performance penalties associated with two-phase commit. Specifically, when propagating a change to a data item from a first system to a second system, the second system compares the first system's 'pre-update' value of the data item with its current value of the data item. If the pre-update value from the first system does not match the current value in the second system, then a conflict has occurred. Upon detecting a conflict, both systems use timestamps associated with the respective conflicting changes to determine which conflicting change 'wins'. The winning change is applied by all systems whose changes did not win.

Mitigating race conditions across two live datastores

A queue-based task management system is provided. Unlike conventional queue-based task management systems, the system described herein does not handle all tasks in the conventional manner. Rather, tasks can be associated with one of several modes, including: a queue-mode (always storing the task in the queue) and an immediate-mode (record the task, but execute it immediately if possible). The two modes may be controlled programmatically to optimize utilization of system resources. Immediate-mode is implemented by monitoring available task system resources, and executing the task immediately in-process if there are resources available; otherwise the task is delegated to the queue. This in-process execution allows the sharing of parent task resources.

Distributed job framework and task queue

Techniques are provided for predicting time remaining for currently-execution jobs. Rather than predict time remaining based on prior executions, time remaining is predicted based on what has happened so far in the current execution. In order to generate predictions for a currently-executing job instance based on statistics about the currently-executing job instance, the system tracks, for each monitored job instance: (a) how many completed-tasks are currently associated with the job instance, and (b) how many created-but-not-completed tasks are associated with the job instance, and then predicts (c) how many not-yet-created tasks the job instance is likely to have.

Automated predictions for not-yet-completed jobs

Growing community of inventors

Fremont, CA, United States of America

Abhijit Karpe