org.corfudb.infrastructure.management

Class CompleteGraphAdvisor

java.lang.Object

org.corfudb.infrastructure.management.CompleteGraphAdvisor

All Implemented Interfaces:

ClusterAdvisor

public class CompleteGraphAdvisor
extends Object
implements ClusterAdvisor

This class is an implementation of ClusterType where the ideal state of the Corfu cluster is a fully connected network (i.e. complete graph) in which there is an active link amongst all nodes in the cluster. Failed and Healed nodes are recommended so that the cluster remains fully connected.

Created by Sam Behnam on 10/27/18.

Constructor Summary

Constructors

Constructor and Description
CompleteGraphAdvisor(String localEndpoint)

Method Summary

Modifier and Type	Method and Description
Optional<NodeRank>	failedServer(ClusterState clusterState, List<String> unresponsiveServers) Provides list of servers from a given layout(epoch) that this implementation of COMPLETE_GRAPH algorithm has determined as failed.
ClusterGraph	getGraph(ClusterState clusterState) Returns a new cluster graph from the cluster state
ClusterType	getType() Get the corresponding ClusterType used in the current instance of ClusterAdvisor.
Optional<NodeRank>	healedServer(ClusterState clusterState, List<String> unresponsiveServers) Provide a server considered to have healed in the Corfu cluster according to the COMPLETE_GRAPH implementation of algorithm for ClusterType.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

CompleteGraphAdvisor

public CompleteGraphAdvisor(@NonNull
                            String localEndpoint)

Method Detail

getType

public ClusterType getType()

Description copied from interface: ClusterAdvisor

Get the corresponding ClusterType used in the current instance of ClusterAdvisor. This strategy represents the characteristics of the underlying algorithm used for making a decision about the failed or healed status of Corfu servers.

Specified by:

getType in interface ClusterAdvisor

Returns:

a concrete instance of ClusterType

failedServer

public Optional<NodeRank> failedServer(ClusterState clusterState,
                                       List<String> unresponsiveServers)

Provides list of servers from a given layout(epoch) that this implementation of COMPLETE_GRAPH algorithm has determined as failed. The implementation of the algorithm in this method is an approach by executing the following steps:

The failed node is the recommendation of this strategy which their removal from cluster will lead to a fully connected cluster.

To find a failed node: - get decision maker - find a failed node. Check if decision maker is not equal to failed node and failed node is not in unresponsive servers already.

Specified by:

failedServer in interface ClusterAdvisor

Parameters:

clusterState - represents the state of connectivity amongst the Corfu cluster nodes from a node's perspective.

unresponsiveServers - list of unresponsive servers.

Returns:

a server considered as failed according to the underlying strategy.

healedServer

public Optional<NodeRank> healedServer(ClusterState clusterState,
                                       List<String> unresponsiveServers)

Provide a server considered to have healed in the Corfu cluster according to the COMPLETE_GRAPH implementation of algorithm for ClusterType.

The node can heal only itself. The node responsible only for itself, can't heal other nodes. It simplifies healing algorithm and guaranties that if node became available it mark itself as a responsible node in the layout. It helps us to simplify analysis/debugging process and brings simple and reliable algorithm for healing process.

Specified by:

healedServer in interface ClusterAdvisor

Parameters:

clusterState - represents the state of connectivity amongst the Corfu cluster nodes from a node's perspective.

unresponsiveServers - unresponsive servers in a layout.

Returns:

a List of servers considered as healed according to the underlying ClusterType.

getGraph

public ClusterGraph getGraph(ClusterState clusterState)

Returns a new cluster graph from the cluster state

Specified by:

getGraph in interface ClusterAdvisor

Parameters:

clusterState - a cluster state

Returns:

a transformed cluster graph