发明名称 DISTRIBUTED STATE MANAGEMENT USING DYNAMIC REPLICATION GRAPHS
摘要 A sequence of configuration-delta messages from a configuration manager of a replication graph established for replicating state information of an application is received at a particular node of the graph. Representations of the messages are stored in a local configuration change repository. Using the repository, the node identifies neighbor source nodes and destination nodes of the graph. Subsequent to receiving a particular configuration-delta message indicating that a neighbor source node has been removed from the graph, a different source node is identified by the particular node, and a request is sent to the different source node to transmit a set of state transition messages pertaining to the application.
申请公布号 US2015379099(A1) 申请公布日期 2015.12.31
申请号 US201414316674 申请日期 2014.06.26
申请人 AMAZON TECHNOLOGIES, INC. 发明人 VERMEULEN ALLAN HENRY;HO CHI HAI;DEARDEUFF MICHAEL BENJAMIN
分类号 G06F17/30 主分类号 G06F17/30
代理机构 代理人
主权项 1. A system, comprising: one or more computing devices configured to: identify, at a first non-committer node of a directed acyclic graph (DAG) of nodes designated to replicate a state of an application, at least a portion of a first replication pathway to a first committer node of the DAG, based at least in part on a first sequence of configuration-delta messages received from a fault-tolerant DAG configuration manager, wherein at least one other node of the DAG has received a different sequence of configuration-delta messages from the fault-tolerant DAG configuration manager;transmit, from the first non-committer node, for replication at respective local storage devices of one or more nodes along the first replication pathway, a state transition message indicating (a) approval of a first candidate state transition requested by a client and (b) a sequence number indicative of an order in which the first candidate state transition is processed at the first non-committer node relative to other candidate state transitions;in response to a determination, by the fault-tolerant DAG configuration manager, that a particular node of the first replication pathway is to be removed from the DAG, initiate a transmission, from the fault-tolerant DAG configuration manager to one or more remaining nodes of the DAG, of a respective configuration-delta message indicating a removal of the particular node from the DAG, without directing the one or more remaining nodes to pause processing of state transitions.
地址 RENO NV US