公开(公告)号：US11789926B2

公开(公告)日：2023-10-17

申请号：US14987649

申请日：2016-01-04

Applicant: NETFLIX, INC.

Inventor： Ioannis Papapanagiotou ,  Shailesh Birari ,  Jason Cacciatore ,  Minh Do ,  Christos Kalantzis

IPC: G06F16/23 ,  G06F16/11 ,  G06F16/174 ,  G06F16/178 ,  G06F11/14 ,  H04L9/40

CPC classification number: G06F16/2365 ,  G06F11/1451 ,  G06F16/128 ,  G06F16/178 ,  G06F16/1744 ,  H04L63/0428 ,  G06F2201/835 ,  G06F2201/84

Abstract: In one embodiment, a data reconciliation engine works with data store nodes included in a distributed data store system to ensure consistency between the data store nodes. In operation, the data reconciliation receives a different data snapshot from each of the data store nodes. In response, the data reconciliation engine generates one or more recommendations designed to resolve inconsistencies between the data snapshots. The data reconciliation engine then transmits each recommendation to a different data store node. Because the data reconciliation engine performs many of the resource-intensive operations included in the data reconciliation process, the resources of the data store nodes may focus primarily on processing client requests instead of performing data reconciliation operations. Consequently, unlike conventional data store node based reconciliation applications, the data reconciliation engine may process large volumes of data without unacceptably increasing the time required for the distributed data store system to respond to client requests.

公开(公告)号：US10749921B2

公开(公告)日：2020-08-18

申请号：US15379299

申请日：2016-12-14

Applicant: NETFLIX Inc.

Inventor： Ioannis Papapanagiotou ,  Shailesh Birari ,  Jason Cacciatore ,  Minh Do ,  Christos Kalantzis

IPC: H04L29/08 ,  G06F11/14 ,  H04L29/06 ,  H04L12/861 ,  G06F16/25

Abstract: In various embodiments, a node manager configures a “new” node as a replacement for an “unavailable” node that was previously included in a distributed data store. First, the node manager identifies a source node that stores client data that was also stored in the unavailable node. Subsequently, the node manager configures the new node to operate as a slave of the source node and streams the client data from the source node to the new node. Finally, the node manager configures the new node to operate as one of multiple masters nodes in the distributed data store. Advantageously, by configuring the node to implement a hybrid of a master-slave replication scheme and a master-master replication scheme, the node manager enables the distributed data store to process client requests without interruption while automatically restoring the previous level of redundancy provided by the distributed data store.