org.corfudb.infrastructure.management

Class ReconfigurationEventHandler

java.lang.Object

org.corfudb.infrastructure.management.ReconfigurationEventHandler

public class ReconfigurationEventHandler
extends Object

The ReconfigurationEventHandler handles the trigger provided by any source or policy detecting a failure or healing in the cluster. It performs the following functions: Cluster recovery: Recovers the cluster on trigger at startup. Handle failures: Handles any failures in the layout based on the desired policy. Handle healing: Handles healing of unresponsive marked nodes which are now responsive. Handle mergeSegments: Handles merging of multiple segments.

Created by zlokhandwala on 11/18/16.

Constructor Summary

Constructors

Constructor and Description
ReconfigurationEventHandler()

Method Summary

Modifier and Type	Method and Description
static boolean	handleFailure(IReconfigurationHandlerPolicy failureHandlerPolicy, Layout currentLayout, CorfuRuntime corfuRuntime, Set<String> failedServers) Takes in the existing layout and a set of failed nodes.
static boolean	handleHealing(CorfuRuntime runtime, Set<String> healedServers, Duration retryQueryTimeout) Takes in the existing layout and a set of healed nodes.
static boolean	handleMergeSegments(CorfuRuntime runtime, Layout layout, Duration retryQueryTimeout) Launches a workflow to restore redundancy and merge all segments.

Methods inherited from class java.lang.Object

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

Constructor Detail

ReconfigurationEventHandler

public ReconfigurationEventHandler()

Method Detail

handleFailure

public static boolean handleFailure(@Nonnull
                                    IReconfigurationHandlerPolicy failureHandlerPolicy,
                                    @Nonnull
                                    Layout currentLayout,
                                    @Nonnull
                                    CorfuRuntime corfuRuntime,
                                    @Nonnull
                                    Set<String> failedServers)

Takes in the existing layout and a set of failed nodes. It first generates a new layout by removing the failed nodes from the existing layout. It then seals the epoch to prevent any client from accessing the stale layout. Finally we run paxos to update all servers with the new layout.

Parameters:

currentLayout - The current layout

corfuRuntime - Connected corfu runtime instance

failedServers - Set of failed server addresses

handleHealing

public static boolean handleHealing(@Nonnull
                                    CorfuRuntime runtime,
                                    @Nonnull
                                    Set<String> healedServers,
                                    @Nonnull
                                    Duration retryQueryTimeout)

Takes in the existing layout and a set of healed nodes. The LayoutManagementView launches a workflow to heal the specified node. This call is blocked until the workflow is completed or aborted. The node is healed and added back to the layout with all components enabled i.e., with its layout, sequencer and log unit server enabled and as an active participant in the data replication view.

Parameters:

runtime - Connected corfu runtime instance

healedServers - Set of healed server addresses

retryQueryTimeout - Timeout to poll for workflow status.

handleMergeSegments

public static boolean handleMergeSegments(@Nonnull
                                          CorfuRuntime runtime,
                                          @Nonnull
                                          Layout layout,
                                          @Nonnull
                                          Duration retryQueryTimeout)

Launches a workflow to restore redundancy and merge all segments. This method sends an orchestrator request to restore redundancy and merge all segments. This task is run asynchronously by the failure detector if it detects multiple segments in the layout. This is a blocking task and will complete if the workflow completes with a success or a failure. To avoid multiple workflow being submitted by different management agents, the management agents need to deterministically provide the same identifiable endpoint to the orchestrator. In this case, we choose the last added endpoint in the first stripe.

Parameters:

runtime - Connected corfu runtime instance.

layout - Latest known layout.

retryQueryTimeout - Timeout to poll for workflow status.

Returns:

True if workflow completed successfully. False otherwise.