opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/shardmanager/ShardManager.java

   1 /*
   2  * Copyright (c) 2014 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   4  * This program and the accompanying materials are made available under the
   5  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   6  * and is available at http://www.eclipse.org/legal/epl-v10.html
   7  */
   8
   9 package org.opendaylight.controller.cluster.datastore.shardmanager;
  10
  11 import static akka.pattern.Patterns.ask;
  12 import static java.util.Objects.requireNonNull;
  13
  14 import akka.actor.ActorRef;
  15 import akka.actor.Address;
  16 import akka.actor.Cancellable;
  17 import akka.actor.OneForOneStrategy;
  18 import akka.actor.PoisonPill;
  19 import akka.actor.Status;
  20 import akka.actor.SupervisorStrategy;
  21 import akka.actor.SupervisorStrategy.Directive;
  22 import akka.cluster.ClusterEvent;
  23 import akka.cluster.ClusterEvent.MemberWeaklyUp;
  24 import akka.cluster.Member;
  25 import akka.dispatch.Futures;
  26 import akka.dispatch.OnComplete;
  27 import akka.japi.Function;
  28 import akka.pattern.Patterns;
  29 import akka.persistence.DeleteSnapshotsFailure;
  30 import akka.persistence.DeleteSnapshotsSuccess;
  31 import akka.persistence.RecoveryCompleted;
  32 import akka.persistence.SaveSnapshotFailure;
  33 import akka.persistence.SaveSnapshotSuccess;
  34 import akka.persistence.SnapshotOffer;
  35 import akka.persistence.SnapshotSelectionCriteria;
  36 import akka.util.Timeout;
  37 import com.google.common.annotations.VisibleForTesting;
  38 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  39 import java.util.ArrayList;
  40 import java.util.Collection;
  41 import java.util.Collections;
  42 import java.util.HashMap;
  43 import java.util.HashSet;
  44 import java.util.List;
  45 import java.util.Map;
  46 import java.util.Map.Entry;
  47 import java.util.Set;
  48 import java.util.concurrent.CountDownLatch;
  49 import java.util.concurrent.TimeUnit;
  50 import java.util.concurrent.TimeoutException;
  51 import java.util.function.Consumer;
  52 import java.util.function.Supplier;
  53 import org.opendaylight.controller.cluster.access.concepts.MemberName;
  54 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActorWithMetering;
  55 import org.opendaylight.controller.cluster.common.actor.Dispatchers;
  56 import org.opendaylight.controller.cluster.datastore.AbstractDataStore;
  57 import org.opendaylight.controller.cluster.datastore.ClusterWrapper;
  58 import org.opendaylight.controller.cluster.datastore.DatastoreContext;
  59 import org.opendaylight.controller.cluster.datastore.DatastoreContext.Builder;
  60 import org.opendaylight.controller.cluster.datastore.DatastoreContextFactory;
  61 import org.opendaylight.controller.cluster.datastore.Shard;
  62 import org.opendaylight.controller.cluster.datastore.config.Configuration;
  63 import org.opendaylight.controller.cluster.datastore.config.ModuleShardConfiguration;
  64 import org.opendaylight.controller.cluster.datastore.config.PrefixShardConfiguration;
  65 import org.opendaylight.controller.cluster.datastore.exceptions.AlreadyExistsException;
  66 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  67 import org.opendaylight.controller.cluster.datastore.exceptions.NotInitializedException;
  68 import org.opendaylight.controller.cluster.datastore.exceptions.PrimaryNotFoundException;
  69 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  70 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  71 import org.opendaylight.controller.cluster.datastore.messages.AddPrefixShardReplica;
  72 import org.opendaylight.controller.cluster.datastore.messages.AddShardReplica;
  73 import org.opendaylight.controller.cluster.datastore.messages.ChangeShardMembersVotingStatus;
  74 import org.opendaylight.controller.cluster.datastore.messages.CreateShard;
  75 import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard;
  76 import org.opendaylight.controller.cluster.datastore.messages.FindPrimary;
  77 import org.opendaylight.controller.cluster.datastore.messages.FlipShardMembersVotingStatus;
  78 import org.opendaylight.controller.cluster.datastore.messages.GetShardRole;
  79 import org.opendaylight.controller.cluster.datastore.messages.GetShardRoleReply;
  80 import org.opendaylight.controller.cluster.datastore.messages.LocalPrimaryShardFound;
  81 import org.opendaylight.controller.cluster.datastore.messages.LocalShardFound;
  82 import org.opendaylight.controller.cluster.datastore.messages.LocalShardNotFound;
  83 import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary;
  84 import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound;
  85 import org.opendaylight.controller.cluster.datastore.messages.RemovePrefixShardReplica;
  86 import org.opendaylight.controller.cluster.datastore.messages.RemoveShardReplica;
  87 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  88 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  89 import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot;
  90 import org.opendaylight.controller.cluster.datastore.persisted.ShardManagerSnapshot;
  91 import org.opendaylight.controller.cluster.datastore.utils.ClusterUtils;
  92 import org.opendaylight.controller.cluster.datastore.utils.CompositeOnComplete;
  93 import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache;
  94 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  95 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListenerReply;
  96 import org.opendaylight.controller.cluster.notifications.RoleChangeNotification;
  97 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  98 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
  99 import org.opendaylight.controller.cluster.raft.client.messages.GetOnDemandRaftState;
 100 import org.opendaylight.controller.cluster.raft.client.messages.GetSnapshot;
 101 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
 102 import org.opendaylight.controller.cluster.raft.client.messages.Shutdown;
 103 import org.opendaylight.controller.cluster.raft.messages.AddServer;
 104 import org.opendaylight.controller.cluster.raft.messages.AddServerReply;
 105 import org.opendaylight.controller.cluster.raft.messages.ChangeServersVotingStatus;
 106 import org.opendaylight.controller.cluster.raft.messages.RemoveServer;
 107 import org.opendaylight.controller.cluster.raft.messages.RemoveServerReply;
 108 import org.opendaylight.controller.cluster.raft.messages.ServerChangeReply;
 109 import org.opendaylight.controller.cluster.raft.messages.ServerChangeStatus;
 110 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
 111 import org.opendaylight.controller.cluster.raft.policy.DisableElectionsRaftPolicy;
 112 import org.opendaylight.controller.cluster.sharding.PrefixedShardConfigUpdateHandler;
 113 import org.opendaylight.controller.cluster.sharding.messages.InitConfigListener;
 114 import org.opendaylight.controller.cluster.sharding.messages.PrefixShardCreated;
 115 import org.opendaylight.controller.cluster.sharding.messages.PrefixShardRemoved;
 116 import org.opendaylight.mdsal.dom.api.DOMDataTreeIdentifier;
 117 import org.opendaylight.yangtools.concepts.Registration;
 118 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
 119 import org.opendaylight.yangtools.yang.model.api.EffectiveModelContext;
 120 import org.slf4j.Logger;
 121 import org.slf4j.LoggerFactory;
 122 import scala.concurrent.ExecutionContext;
 123 import scala.concurrent.Future;
 124 import scala.concurrent.duration.FiniteDuration;
 125
 126 /**
 127  * Manages the shards for a data store. The ShardManager has the following jobs:
 128  * <ul>
 129  * <li> Create all the local shard replicas that belong on this cluster member
 130  * <li> Find the address of the local shard
 131  * <li> Find the primary replica for any given shard
 132  * <li> Monitor the cluster members and store their addresses
 133  * </ul>
 134  */
 135 class ShardManager extends AbstractUntypedPersistentActorWithMetering {
 136     private static final Logger LOG = LoggerFactory.getLogger(ShardManager.class);
 137
 138     // Stores a mapping between a shard name and it's corresponding information
 139     // Shard names look like inventory, topology etc and are as specified in
 140     // configuration
 141     @VisibleForTesting
 142     final Map<String, ShardInformation> localShards = new HashMap<>();
 143
 144     // The type of a ShardManager reflects the type of the datastore itself
 145     // A data store could be of type config/operational
 146     private final String type;
 147
 148     private final ClusterWrapper cluster;
 149
 150     private final Configuration configuration;
 151
 152     @VisibleForTesting
 153     final String shardDispatcherPath;
 154
 155     private final ShardManagerInfo shardManagerMBean;
 156
 157     private DatastoreContextFactory datastoreContextFactory;
 158
 159     private final CountDownLatch waitTillReadyCountdownLatch;
 160
 161     private final PrimaryShardInfoFutureCache primaryShardInfoCache;
 162
 163     @VisibleForTesting
 164     final ShardPeerAddressResolver peerAddressResolver;
 165
 166     private EffectiveModelContext schemaContext;
 167
 168     private DatastoreSnapshot restoreFromSnapshot;
 169
 170     private ShardManagerSnapshot currentSnapshot;
 171
 172     private final Set<String> shardReplicaOperationsInProgress = new HashSet<>();
 173
 174     private final Map<String, CompositeOnComplete<Boolean>> shardActorsStopping = new HashMap<>();
 175
 176     private final Set<Consumer<String>> shardAvailabilityCallbacks = new HashSet<>();
 177
 178     private final String persistenceId;
 179     private final AbstractDataStore dataStore;
 180
 181     private PrefixedShardConfigUpdateHandler configUpdateHandler;
 182
 183     ShardManager(final AbstractShardManagerCreator<?> builder) {
 184         this.cluster = builder.getCluster();
 185         this.configuration = builder.getConfiguration();
 186         this.datastoreContextFactory = builder.getDatastoreContextFactory();
 187         this.type = datastoreContextFactory.getBaseDatastoreContext().getDataStoreName();
 188         this.shardDispatcherPath =
 189                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Shard);
 190         this.waitTillReadyCountdownLatch = builder.getWaitTillReadyCountDownLatch();
 191         this.primaryShardInfoCache = builder.getPrimaryShardInfoCache();
 192         this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
 193
 194         String possiblePersistenceId = datastoreContextFactory.getBaseDatastoreContext().getShardManagerPersistenceId();
 195         persistenceId = possiblePersistenceId != null ? possiblePersistenceId : "shard-manager-" + type;
 196
 197         peerAddressResolver = new ShardPeerAddressResolver(type, cluster.getCurrentMemberName());
 198
 199         // Subscribe this actor to cluster member events
 200         cluster.subscribeToMemberEvents(getSelf());
 201
 202         shardManagerMBean = new ShardManagerInfo(getSelf(), cluster.getCurrentMemberName(),
 203                 "shard-manager-" + this.type,
 204                 datastoreContextFactory.getBaseDatastoreContext().getDataStoreMXBeanType());
 205         shardManagerMBean.registerMBean();
 206
 207         dataStore = builder.getDistributedDataStore();
 208     }
 209
 210     @Override
 211     public void preStart() {
 212         LOG.info("Starting ShardManager {}", persistenceId);
 213     }
 214
 215     @Override
 216     public void postStop() {
 217         LOG.info("Stopping ShardManager {}", persistenceId());
 218
 219         shardManagerMBean.unregisterMBean();
 220     }
 221
 222     @Override
 223     public void handleCommand(final Object message) throws Exception {
 224         if (message  instanceof FindPrimary) {
 225             findPrimary((FindPrimary)message);
 226         } else if (message instanceof FindLocalShard) {
 227             findLocalShard((FindLocalShard) message);
 228         } else if (message instanceof UpdateSchemaContext) {
 229             updateSchemaContext(message);
 230         } else if (message instanceof ActorInitialized) {
 231             onActorInitialized(message);
 232         } else if (message instanceof ClusterEvent.MemberUp) {
 233             memberUp((ClusterEvent.MemberUp) message);
 234         } else if (message instanceof ClusterEvent.MemberWeaklyUp) {
 235             memberWeaklyUp((ClusterEvent.MemberWeaklyUp) message);
 236         } else if (message instanceof ClusterEvent.MemberExited) {
 237             memberExited((ClusterEvent.MemberExited) message);
 238         } else if (message instanceof ClusterEvent.MemberRemoved) {
 239             memberRemoved((ClusterEvent.MemberRemoved) message);
 240         } else if (message instanceof ClusterEvent.UnreachableMember) {
 241             memberUnreachable((ClusterEvent.UnreachableMember) message);
 242         } else if (message instanceof ClusterEvent.ReachableMember) {
 243             memberReachable((ClusterEvent.ReachableMember) message);
 244         } else if (message instanceof DatastoreContextFactory) {
 245             onDatastoreContextFactory((DatastoreContextFactory) message);
 246         } else if (message instanceof RoleChangeNotification) {
 247             onRoleChangeNotification((RoleChangeNotification) message);
 248         } else if (message instanceof FollowerInitialSyncUpStatus) {
 249             onFollowerInitialSyncStatus((FollowerInitialSyncUpStatus) message);
 250         } else if (message instanceof ShardNotInitializedTimeout) {
 251             onShardNotInitializedTimeout((ShardNotInitializedTimeout) message);
 252         } else if (message instanceof ShardLeaderStateChanged) {
 253             onLeaderStateChanged((ShardLeaderStateChanged) message);
 254         } else if (message instanceof SwitchShardBehavior) {
 255             onSwitchShardBehavior((SwitchShardBehavior) message);
 256         } else if (message instanceof CreateShard) {
 257             onCreateShard((CreateShard)message);
 258         } else if (message instanceof AddShardReplica) {
 259             onAddShardReplica((AddShardReplica) message);
 260         } else if (message instanceof AddPrefixShardReplica) {
 261             onAddPrefixShardReplica((AddPrefixShardReplica) message);
 262         } else if (message instanceof PrefixShardCreated) {
 263             onPrefixShardCreated((PrefixShardCreated) message);
 264         } else if (message instanceof PrefixShardRemoved) {
 265             onPrefixShardRemoved((PrefixShardRemoved) message);
 266         } else if (message instanceof InitConfigListener) {
 267             onInitConfigListener();
 268         } else if (message instanceof ForwardedAddServerReply) {
 269             ForwardedAddServerReply msg = (ForwardedAddServerReply)message;
 270             onAddServerReply(msg.shardInfo, msg.addServerReply, getSender(), msg.leaderPath,
 271                     msg.removeShardOnFailure);
 272         } else if (message instanceof ForwardedAddServerFailure) {
 273             ForwardedAddServerFailure msg = (ForwardedAddServerFailure)message;
 274             onAddServerFailure(msg.shardName, msg.failureMessage, msg.failure, getSender(), msg.removeShardOnFailure);
 275         } else if (message instanceof RemoveShardReplica) {
 276             onRemoveShardReplica((RemoveShardReplica) message);
 277         } else if (message instanceof RemovePrefixShardReplica) {
 278             onRemovePrefixShardReplica((RemovePrefixShardReplica) message);
 279         } else if (message instanceof WrappedShardResponse) {
 280             onWrappedShardResponse((WrappedShardResponse) message);
 281         } else if (message instanceof GetSnapshot) {
 282             onGetSnapshot((GetSnapshot) message);
 283         } else if (message instanceof ServerRemoved) {
 284             onShardReplicaRemoved((ServerRemoved) message);
 285         } else if (message instanceof ChangeShardMembersVotingStatus) {
 286             onChangeShardServersVotingStatus((ChangeShardMembersVotingStatus) message);
 287         } else if (message instanceof FlipShardMembersVotingStatus) {
 288             onFlipShardMembersVotingStatus((FlipShardMembersVotingStatus) message);
 289         } else if (message instanceof SaveSnapshotSuccess) {
 290             onSaveSnapshotSuccess((SaveSnapshotSuccess) message);
 291         } else if (message instanceof SaveSnapshotFailure) {
 292             LOG.error("{}: SaveSnapshotFailure received for saving snapshot of shards", persistenceId(),
 293                     ((SaveSnapshotFailure) message).cause());
 294         } else if (message instanceof Shutdown) {
 295             onShutDown();
 296         } else if (message instanceof GetLocalShardIds) {
 297             onGetLocalShardIds();
 298         } else if (message instanceof GetShardRole) {
 299             onGetShardRole((GetShardRole) message);
 300         } else if (message instanceof RunnableMessage) {
 301             ((RunnableMessage)message).run();
 302         } else if (message instanceof RegisterForShardAvailabilityChanges) {
 303             onRegisterForShardAvailabilityChanges((RegisterForShardAvailabilityChanges)message);
 304         } else if (message instanceof DeleteSnapshotsFailure) {
 305             LOG.warn("{}: Failed to delete prior snapshots", persistenceId(),
 306                     ((DeleteSnapshotsFailure) message).cause());
 307         } else if (message instanceof DeleteSnapshotsSuccess) {
 308             LOG.debug("{}: Successfully deleted prior snapshots", persistenceId());
 309         } else if (message instanceof RegisterRoleChangeListenerReply) {
 310             LOG.trace("{}: Received RegisterRoleChangeListenerReply", persistenceId());
 311         } else if (message instanceof ClusterEvent.MemberEvent) {
 312             LOG.trace("{}: Received other ClusterEvent.MemberEvent: {}", persistenceId(), message);
 313         } else {
 314             unknownMessage(message);
 315         }
 316     }
 317
 318     private void onRegisterForShardAvailabilityChanges(final RegisterForShardAvailabilityChanges message) {
 319         LOG.debug("{}: onRegisterForShardAvailabilityChanges: {}", persistenceId(), message);
 320
 321         final Consumer<String> callback = message.getCallback();
 322         shardAvailabilityCallbacks.add(callback);
 323
 324         getSender().tell(new Status.Success((Registration)
 325             () -> executeInSelf(() -> shardAvailabilityCallbacks.remove(callback))), self());
 326     }
 327
 328     private void onGetShardRole(final GetShardRole message) {
 329         LOG.debug("{}: onGetShardRole for shard: {}", persistenceId(), message.getName());
 330
 331         final String name = message.getName();
 332
 333         final ShardInformation shardInformation = localShards.get(name);
 334
 335         if (shardInformation == null) {
 336             LOG.info("{}: no shard information for {} found", persistenceId(), name);
 337             getSender().tell(new Status.Failure(
 338                     new IllegalArgumentException("Shard with name " + name + " not present.")), ActorRef.noSender());
 339             return;
 340         }
 341
 342         getSender().tell(new GetShardRoleReply(shardInformation.getRole()), ActorRef.noSender());
 343     }
 344
 345     private void onInitConfigListener() {
 346         LOG.debug("{}: Initializing config listener on {}", persistenceId(), cluster.getCurrentMemberName());
 347
 348         final org.opendaylight.mdsal.common.api.LogicalDatastoreType datastoreType =
 349                 org.opendaylight.mdsal.common.api.LogicalDatastoreType
 350                         .valueOf(datastoreContextFactory.getBaseDatastoreContext().getLogicalStoreType().name());
 351
 352         if (configUpdateHandler != null) {
 353             configUpdateHandler.close();
 354         }
 355
 356         configUpdateHandler = new PrefixedShardConfigUpdateHandler(self(), cluster.getCurrentMemberName());
 357         configUpdateHandler.initListener(dataStore, datastoreType);
 358     }
 359
 360     void onShutDown() {
 361         List<Future<Boolean>> stopFutures = new ArrayList<>(localShards.size());
 362         for (ShardInformation info : localShards.values()) {
 363             if (info.getActor() != null) {
 364                 LOG.debug("{}: Issuing gracefulStop to shard {}", persistenceId(), info.getShardId());
 365
 366                 FiniteDuration duration = info.getDatastoreContext().getShardRaftConfig()
 367                         .getElectionTimeOutInterval().$times(2);
 368                 stopFutures.add(Patterns.gracefulStop(info.getActor(), duration, Shutdown.INSTANCE));
 369             }
 370         }
 371
 372         LOG.info("Shutting down ShardManager {} - waiting on {} shards", persistenceId(), stopFutures.size());
 373
 374         ExecutionContext dispatcher = new Dispatchers(context().system().dispatchers())
 375                 .getDispatcher(Dispatchers.DispatcherType.Client);
 376         Future<Iterable<Boolean>> combinedFutures = Futures.sequence(stopFutures, dispatcher);
 377
 378         combinedFutures.onComplete(new OnComplete<Iterable<Boolean>>() {
 379             @Override
 380             public void onComplete(final Throwable failure, final Iterable<Boolean> results) {
 381                 LOG.debug("{}: All shards shutdown - sending PoisonPill to self", persistenceId());
 382
 383                 self().tell(PoisonPill.getInstance(), self());
 384
 385                 if (failure != null) {
 386                     LOG.warn("{}: An error occurred attempting to shut down the shards", persistenceId(), failure);
 387                 } else {
 388                     int nfailed = 0;
 389                     for (Boolean result : results) {
 390                         if (!result) {
 391                             nfailed++;
 392                         }
 393                     }
 394
 395                     if (nfailed > 0) {
 396                         LOG.warn("{}: {} shards did not shut down gracefully", persistenceId(), nfailed);
 397                     }
 398                 }
 399             }
 400         }, dispatcher);
 401     }
 402
 403     private void onWrappedShardResponse(final WrappedShardResponse message) {
 404         if (message.getResponse() instanceof RemoveServerReply) {
 405             onRemoveServerReply(getSender(), message.getShardId(), (RemoveServerReply) message.getResponse(),
 406                     message.getLeaderPath());
 407         }
 408     }
 409
 410     private void onRemoveServerReply(final ActorRef originalSender, final ShardIdentifier shardId,
 411             final RemoveServerReply replyMsg, final String leaderPath) {
 412         shardReplicaOperationsInProgress.remove(shardId.getShardName());
 413
 414         LOG.debug("{}: Received {} for shard {}", persistenceId(), replyMsg, shardId.getShardName());
 415
 416         if (replyMsg.getStatus() == ServerChangeStatus.OK) {
 417             LOG.debug("{}: Leader shard successfully removed the replica shard {}", persistenceId(),
 418                     shardId.getShardName());
 419             originalSender.tell(new Status.Success(null), getSelf());
 420         } else {
 421             LOG.warn("{}: Leader failed to remove shard replica {} with status {}",
 422                     persistenceId(), shardId, replyMsg.getStatus());
 423
 424             Exception failure = getServerChangeException(RemoveServer.class, replyMsg.getStatus(), leaderPath, shardId);
 425             originalSender.tell(new Status.Failure(failure), getSelf());
 426         }
 427     }
 428
 429     @SuppressFBWarnings(value = "UPM_UNCALLED_PRIVATE_METHOD",
 430             justification = "https://github.com/spotbugs/spotbugs/issues/811")
 431     private void removePrefixShardReplica(final RemovePrefixShardReplica contextMessage, final String shardName,
 432                                           final String primaryPath, final ActorRef sender) {
 433         if (isShardReplicaOperationInProgress(shardName, sender)) {
 434             return;
 435         }
 436
 437         shardReplicaOperationsInProgress.add(shardName);
 438
 439         final ShardIdentifier shardId = getShardIdentifier(contextMessage.getMemberName(), shardName);
 440
 441         final DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
 442
 443         //inform ShardLeader to remove this shard as a replica by sending an RemoveServer message
 444         LOG.debug("{}: Sending RemoveServer message to peer {} for shard {}", persistenceId(),
 445                 primaryPath, shardId);
 446
 447         Timeout removeServerTimeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().duration());
 448         Future<Object> futureObj = ask(getContext().actorSelection(primaryPath),
 449                 new RemoveServer(shardId.toString()), removeServerTimeout);
 450
 451         futureObj.onComplete(new OnComplete<>() {
 452             @Override
 453             public void onComplete(final Throwable failure, final Object response) {
 454                 if (failure != null) {
 455                     shardReplicaOperationsInProgress.remove(shardName);
 456
 457                     LOG.debug("{}: RemoveServer request to leader {} for shard {} failed", persistenceId(), primaryPath,
 458                         shardName, failure);
 459
 460                     // FAILURE
 461                     sender.tell(new Status.Failure(new RuntimeException(
 462                         String.format("RemoveServer request to leader %s for shard %s failed", primaryPath, shardName),
 463                         failure)), self());
 464                 } else {
 465                     // SUCCESS
 466                     self().tell(new WrappedShardResponse(shardId, response, primaryPath), sender);
 467                 }
 468             }
 469         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
 470     }
 471
 472     @SuppressFBWarnings(value = "UPM_UNCALLED_PRIVATE_METHOD",
 473             justification = "https://github.com/spotbugs/spotbugs/issues/811")
 474     private void removeShardReplica(final RemoveShardReplica contextMessage, final String shardName,
 475             final String primaryPath, final ActorRef sender) {
 476         if (isShardReplicaOperationInProgress(shardName, sender)) {
 477             return;
 478         }
 479
 480         shardReplicaOperationsInProgress.add(shardName);
 481
 482         final ShardIdentifier shardId = getShardIdentifier(contextMessage.getMemberName(), shardName);
 483
 484         final DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
 485
 486         //inform ShardLeader to remove this shard as a replica by sending an RemoveServer message
 487         LOG.debug("{}: Sending RemoveServer message to peer {} for shard {}", persistenceId(),
 488                 primaryPath, shardId);
 489
 490         Timeout removeServerTimeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().duration());
 491         Future<Object> futureObj = ask(getContext().actorSelection(primaryPath),
 492                 new RemoveServer(shardId.toString()), removeServerTimeout);
 493
 494         futureObj.onComplete(new OnComplete<>() {
 495             @Override
 496             public void onComplete(final Throwable failure, final Object response) {
 497                 if (failure != null) {
 498                     shardReplicaOperationsInProgress.remove(shardName);
 499                     LOG.debug("{}: RemoveServer request to leader {} for shard {} failed", persistenceId(), primaryPath,
 500                         shardName, failure);
 501
 502                     // FAILURE
 503                     sender.tell(new Status.Failure(new RuntimeException(
 504                         String.format("RemoveServer request to leader %s for shard %s failed", primaryPath, shardName),
 505                         failure)), self());
 506                 } else {
 507                     // SUCCESS
 508                     self().tell(new WrappedShardResponse(shardId, response, primaryPath), sender);
 509                 }
 510             }
 511         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
 512     }
 513
 514     private void onShardReplicaRemoved(final ServerRemoved message) {
 515         removeShard(new ShardIdentifier.Builder().fromShardIdString(message.getServerId()).build());
 516     }
 517
 518     @SuppressWarnings("checkstyle:IllegalCatch")
 519     private void removeShard(final ShardIdentifier shardId) {
 520         final String shardName = shardId.getShardName();
 521         final ShardInformation shardInformation = localShards.remove(shardName);
 522         if (shardInformation == null) {
 523             LOG.debug("{} : Shard replica {} is not present in list", persistenceId(), shardId.toString());
 524             return;
 525         }
 526
 527         final ActorRef shardActor = shardInformation.getActor();
 528         if (shardActor != null) {
 529             long timeoutInMS = Math.max(shardInformation.getDatastoreContext().getShardRaftConfig()
 530                     .getElectionTimeOutInterval().$times(3).toMillis(), 10000);
 531
 532             LOG.debug("{} : Sending Shutdown to Shard actor {} with {} ms timeout", persistenceId(), shardActor,
 533                     timeoutInMS);
 534
 535             final Future<Boolean> stopFuture = Patterns.gracefulStop(shardActor,
 536                     FiniteDuration.apply(timeoutInMS, TimeUnit.MILLISECONDS), Shutdown.INSTANCE);
 537
 538             final CompositeOnComplete<Boolean> onComplete = new CompositeOnComplete<>() {
 539                 @Override
 540                 public void onComplete(final Throwable failure, final Boolean result) {
 541                     if (failure == null) {
 542                         LOG.debug("{} : Successfully shut down Shard actor {}", persistenceId(), shardActor);
 543                     } else {
 544                         LOG.warn("{}: Failed to shut down Shard actor {}", persistenceId(), shardActor, failure);
 545                     }
 546
 547                     self().tell((RunnableMessage) () -> {
 548                         // At any rate, invalidate primaryShardInfo cache
 549                         primaryShardInfoCache.remove(shardName);
 550
 551                         shardActorsStopping.remove(shardName);
 552                         notifyOnCompleteTasks(failure, result);
 553                     }, ActorRef.noSender());
 554                 }
 555             };
 556
 557             shardActorsStopping.put(shardName, onComplete);
 558             stopFuture.onComplete(onComplete, new Dispatchers(context().system().dispatchers())
 559                     .getDispatcher(Dispatchers.DispatcherType.Client));
 560         }
 561
 562         LOG.debug("{} : Local Shard replica for shard {} has been removed", persistenceId(), shardName);
 563         persistShardList();
 564     }
 565
 566     private void onGetSnapshot(final GetSnapshot getSnapshot) {
 567         LOG.debug("{}: onGetSnapshot", persistenceId());
 568
 569         List<String> notInitialized = null;
 570         for (ShardInformation shardInfo : localShards.values()) {
 571             if (!shardInfo.isShardInitialized()) {
 572                 if (notInitialized == null) {
 573                     notInitialized = new ArrayList<>();
 574                 }
 575
 576                 notInitialized.add(shardInfo.getShardName());
 577             }
 578         }
 579
 580         if (notInitialized != null) {
 581             getSender().tell(new Status.Failure(new IllegalStateException(String.format(
 582                     "%d shard(s) %s are not initialized", notInitialized.size(), notInitialized))), getSelf());
 583             return;
 584         }
 585
 586         ActorRef replyActor = getContext().actorOf(ShardManagerGetSnapshotReplyActor.props(
 587                 new ArrayList<>(localShards.keySet()), type, currentSnapshot , getSender(), persistenceId(),
 588                 datastoreContextFactory.getBaseDatastoreContext().getShardInitializationTimeout().duration()));
 589
 590         for (ShardInformation shardInfo: localShards.values()) {
 591             shardInfo.getActor().tell(getSnapshot, replyActor);
 592         }
 593     }
 594
 595     @SuppressWarnings("checkstyle:IllegalCatch")
 596     private void onCreateShard(final CreateShard createShard) {
 597         LOG.debug("{}: onCreateShard: {}", persistenceId(), createShard);
 598
 599         Object reply;
 600         try {
 601             String shardName = createShard.getModuleShardConfig().getShardName();
 602             if (localShards.containsKey(shardName)) {
 603                 LOG.debug("{}: Shard {} already exists", persistenceId(), shardName);
 604                 reply = new Status.Success(String.format("Shard with name %s already exists", shardName));
 605             } else {
 606                 doCreateShard(createShard);
 607                 reply = new Status.Success(null);
 608             }
 609         } catch (Exception e) {
 610             LOG.error("{}: onCreateShard failed", persistenceId(), e);
 611             reply = new Status.Failure(e);
 612         }
 613
 614         if (getSender() != null && !getContext().system().deadLetters().equals(getSender())) {
 615             getSender().tell(reply, getSelf());
 616         }
 617     }
 618
 619     private void onPrefixShardCreated(final PrefixShardCreated message) {
 620         LOG.debug("{}: onPrefixShardCreated: {}", persistenceId(), message);
 621
 622         final PrefixShardConfiguration config = message.getConfiguration();
 623         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
 624                 ClusterUtils.getCleanShardName(config.getPrefix().getRootIdentifier()));
 625         final String shardName = shardId.getShardName();
 626
 627         if (isPreviousShardActorStopInProgress(shardName, message)) {
 628             return;
 629         }
 630
 631         if (localShards.containsKey(shardName)) {
 632             LOG.debug("{}: Received create for an already existing shard {}", persistenceId(), shardName);
 633             final PrefixShardConfiguration existing =
 634                     configuration.getAllPrefixShardConfigurations().get(config.getPrefix());
 635
 636             if (existing != null && existing.equals(config)) {
 637                 // we don't have to do nothing here
 638                 return;
 639             }
 640         }
 641
 642         doCreatePrefixShard(config, shardId, shardName);
 643     }
 644
 645     private boolean isPreviousShardActorStopInProgress(final String shardName, final Object messageToDefer) {
 646         final CompositeOnComplete<Boolean> stopOnComplete = shardActorsStopping.get(shardName);
 647         if (stopOnComplete == null) {
 648             return false;
 649         }
 650
 651         LOG.debug("{} : Stop is in progress for shard {} - adding OnComplete callback to defer {}", persistenceId(),
 652                 shardName, messageToDefer);
 653         final ActorRef sender = getSender();
 654         stopOnComplete.addOnComplete(new OnComplete<Boolean>() {
 655             @Override
 656             public void onComplete(final Throwable failure, final Boolean result) {
 657                 LOG.debug("{} : Stop complete for shard {} - re-queing {}", persistenceId(), shardName, messageToDefer);
 658                 self().tell(messageToDefer, sender);
 659             }
 660         });
 661
 662         return true;
 663     }
 664
 665     private void doCreatePrefixShard(final PrefixShardConfiguration config, final ShardIdentifier shardId,
 666             final String shardName) {
 667         configuration.addPrefixShardConfiguration(config);
 668
 669         final Builder builder = newShardDatastoreContextBuilder(shardName);
 670         builder.logicalStoreType(config.getPrefix().getDatastoreType())
 671                 .storeRoot(config.getPrefix().getRootIdentifier());
 672         DatastoreContext shardDatastoreContext = builder.build();
 673
 674         final Map<String, String> peerAddresses = getPeerAddresses(shardName);
 675         final boolean isActiveMember = true;
 676
 677         LOG.debug("{} doCreatePrefixShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
 678                 persistenceId(), shardId, config.getShardMemberNames(), peerAddresses, isActiveMember);
 679
 680         final ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
 681                 shardDatastoreContext, Shard.builder(), peerAddressResolver);
 682         info.setActiveMember(isActiveMember);
 683         localShards.put(info.getShardName(), info);
 684
 685         if (schemaContext != null) {
 686             info.setSchemaContext(schemaContext);
 687             info.setActor(newShardActor(info));
 688         }
 689     }
 690
 691     private void onPrefixShardRemoved(final PrefixShardRemoved message) {
 692         LOG.debug("{}: onPrefixShardRemoved : {}", persistenceId(), message);
 693
 694         final DOMDataTreeIdentifier prefix = message.getPrefix();
 695         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
 696                 ClusterUtils.getCleanShardName(prefix.getRootIdentifier()));
 697
 698         configuration.removePrefixShardConfiguration(prefix);
 699         removeShard(shardId);
 700     }
 701
 702     private void doCreateShard(final CreateShard createShard) {
 703         final ModuleShardConfiguration moduleShardConfig = createShard.getModuleShardConfig();
 704         final String shardName = moduleShardConfig.getShardName();
 705
 706         configuration.addModuleShardConfiguration(moduleShardConfig);
 707
 708         DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
 709         if (shardDatastoreContext == null) {
 710             shardDatastoreContext = newShardDatastoreContext(shardName);
 711         } else {
 712             shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).shardPeerAddressResolver(
 713                     peerAddressResolver).build();
 714         }
 715
 716         ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
 717
 718         boolean shardWasInRecoveredSnapshot = currentSnapshot != null
 719                 && currentSnapshot.getShardList().contains(shardName);
 720
 721         Map<String, String> peerAddresses;
 722         boolean isActiveMember;
 723         if (shardWasInRecoveredSnapshot || configuration.getMembersFromShardName(shardName)
 724                 .contains(cluster.getCurrentMemberName())) {
 725             peerAddresses = getPeerAddresses(shardName);
 726             isActiveMember = true;
 727         } else {
 728             // The local member is not in the static shard member configuration and the shard did not
 729             // previously exist (ie !shardWasInRecoveredSnapshot). In this case we'll create
 730             // the shard with no peers and with elections disabled so it stays as follower. A
 731             // subsequent AddServer request will be needed to make it an active member.
 732             isActiveMember = false;
 733             peerAddresses = Collections.emptyMap();
 734             shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext)
 735                     .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
 736         }
 737
 738         LOG.debug("{} doCreateShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
 739                 persistenceId(), shardId, moduleShardConfig.getShardMemberNames(), peerAddresses,
 740                 isActiveMember);
 741
 742         ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
 743                 shardDatastoreContext, createShard.getShardBuilder(), peerAddressResolver);
 744         info.setActiveMember(isActiveMember);
 745         localShards.put(info.getShardName(), info);
 746
 747         if (schemaContext != null) {
 748             info.setSchemaContext(schemaContext);
 749             info.setActor(newShardActor(info));
 750         }
 751     }
 752
 753     private DatastoreContext.Builder newShardDatastoreContextBuilder(final String shardName) {
 754         return DatastoreContext.newBuilderFrom(datastoreContextFactory.getShardDatastoreContext(shardName))
 755                 .shardPeerAddressResolver(peerAddressResolver);
 756     }
 757
 758     private DatastoreContext newShardDatastoreContext(final String shardName) {
 759         return newShardDatastoreContextBuilder(shardName).build();
 760     }
 761
 762     private void checkReady() {
 763         if (isReadyWithLeaderId()) {
 764             LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
 765                     persistenceId(), type, waitTillReadyCountdownLatch.getCount());
 766
 767             waitTillReadyCountdownLatch.countDown();
 768         }
 769     }
 770
 771     private void onLeaderStateChanged(final ShardLeaderStateChanged leaderStateChanged) {
 772         LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
 773
 774         ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
 775         if (shardInformation != null) {
 776             shardInformation.setLocalDataTree(leaderStateChanged.getLocalShardDataTree());
 777             shardInformation.setLeaderVersion(leaderStateChanged.getLeaderPayloadVersion());
 778             if (shardInformation.setLeaderId(leaderStateChanged.getLeaderId())) {
 779                 primaryShardInfoCache.remove(shardInformation.getShardName());
 780
 781                 notifyShardAvailabilityCallbacks(shardInformation);
 782             }
 783
 784             checkReady();
 785         } else {
 786             LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
 787         }
 788     }
 789
 790     private void notifyShardAvailabilityCallbacks(final ShardInformation shardInformation) {
 791         shardAvailabilityCallbacks.forEach(callback -> callback.accept(shardInformation.getShardName()));
 792     }
 793
 794     private void onShardNotInitializedTimeout(final ShardNotInitializedTimeout message) {
 795         ShardInformation shardInfo = message.getShardInfo();
 796
 797         LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
 798                 shardInfo.getShardName());
 799
 800         shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
 801
 802         if (!shardInfo.isShardInitialized()) {
 803             LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
 804             message.getSender().tell(createNotInitializedException(shardInfo.getShardId()), getSelf());
 805         } else {
 806             LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
 807             message.getSender().tell(new NoShardLeaderException(shardInfo.getShardId()), getSelf());
 808         }
 809     }
 810
 811     private void onFollowerInitialSyncStatus(final FollowerInitialSyncUpStatus status) {
 812         LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
 813                 status.getName(), status.isInitialSyncDone());
 814
 815         ShardInformation shardInformation = findShardInformation(status.getName());
 816
 817         if (shardInformation != null) {
 818             shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
 819
 820             shardManagerMBean.setSyncStatus(isInSync());
 821         }
 822
 823     }
 824
 825     private void onRoleChangeNotification(final RoleChangeNotification roleChanged) {
 826         LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
 827                 roleChanged.getOldRole(), roleChanged.getNewRole());
 828
 829         ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
 830         if (shardInformation != null) {
 831             shardInformation.setRole(roleChanged.getNewRole());
 832             checkReady();
 833             shardManagerMBean.setSyncStatus(isInSync());
 834         }
 835     }
 836
 837
 838     private ShardInformation findShardInformation(final String memberId) {
 839         for (ShardInformation info : localShards.values()) {
 840             if (info.getShardId().toString().equals(memberId)) {
 841                 return info;
 842             }
 843         }
 844
 845         return null;
 846     }
 847
 848     private boolean isReadyWithLeaderId() {
 849         boolean isReady = true;
 850         for (ShardInformation info : localShards.values()) {
 851             if (!info.isShardReadyWithLeaderId()) {
 852                 isReady = false;
 853                 break;
 854             }
 855         }
 856         return isReady;
 857     }
 858
 859     private boolean isInSync() {
 860         for (ShardInformation info : localShards.values()) {
 861             if (!info.isInSync()) {
 862                 return false;
 863             }
 864         }
 865         return true;
 866     }
 867
 868     private void onActorInitialized(final Object message) {
 869         final ActorRef sender = getSender();
 870
 871         if (sender == null) {
 872             // why is a non-actor sending this message? Just ignore.
 873             return;
 874         }
 875
 876         String actorName = sender.path().name();
 877         //find shard name from actor name; actor name is stringified shardId
 878
 879         final ShardIdentifier shardId;
 880         try {
 881             shardId = ShardIdentifier.fromShardIdString(actorName);
 882         } catch (IllegalArgumentException e) {
 883             LOG.debug("{}: ignoring actor {}", persistenceId, actorName, e);
 884             return;
 885         }
 886
 887         markShardAsInitialized(shardId.getShardName());
 888     }
 889
 890     private void markShardAsInitialized(final String shardName) {
 891         LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
 892
 893         ShardInformation shardInformation = localShards.get(shardName);
 894         if (shardInformation != null) {
 895             shardInformation.setActorInitialized();
 896
 897             shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
 898         }
 899     }
 900
 901     @Override
 902     protected void handleRecover(final Object message) throws Exception {
 903         if (message instanceof RecoveryCompleted) {
 904             onRecoveryCompleted();
 905         } else if (message instanceof SnapshotOffer) {
 906             applyShardManagerSnapshot((ShardManagerSnapshot)((SnapshotOffer) message).snapshot());
 907         }
 908     }
 909
 910     @SuppressWarnings("checkstyle:IllegalCatch")
 911     private void onRecoveryCompleted() {
 912         LOG.info("Recovery complete : {}", persistenceId());
 913
 914         if (currentSnapshot == null && restoreFromSnapshot != null
 915                 && restoreFromSnapshot.getShardManagerSnapshot() != null) {
 916             ShardManagerSnapshot snapshot = restoreFromSnapshot.getShardManagerSnapshot();
 917
 918             LOG.debug("{}: Restoring from ShardManagerSnapshot: {}", persistenceId(), snapshot);
 919
 920             applyShardManagerSnapshot(snapshot);
 921         }
 922
 923         createLocalShards();
 924     }
 925
 926     private void sendResponse(final ShardInformation shardInformation, final boolean doWait,
 927             final boolean wantShardReady, final Supplier<Object> messageSupplier) {
 928         if (!shardInformation.isShardInitialized() || wantShardReady && !shardInformation.isShardReadyWithLeaderId()) {
 929             if (doWait) {
 930                 final ActorRef sender = getSender();
 931                 final ActorRef self = self();
 932
 933                 Runnable replyRunnable = () -> sender.tell(messageSupplier.get(), self);
 934
 935                 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
 936                     new OnShardInitialized(replyRunnable);
 937
 938                 shardInformation.addOnShardInitialized(onShardInitialized);
 939
 940                 FiniteDuration timeout = shardInformation.getDatastoreContext()
 941                         .getShardInitializationTimeout().duration();
 942                 if (shardInformation.isShardInitialized()) {
 943                     // If the shard is already initialized then we'll wait enough time for the shard to
 944                     // elect a leader, ie 2 times the election timeout.
 945                     timeout = FiniteDuration.create(shardInformation.getDatastoreContext().getShardRaftConfig()
 946                             .getElectionTimeOutInterval().toMillis() * 2, TimeUnit.MILLISECONDS);
 947                 }
 948
 949                 LOG.debug("{}: Scheduling {} ms timer to wait for shard {}", persistenceId(), timeout.toMillis(),
 950                         shardInformation);
 951
 952                 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
 953                         timeout, getSelf(),
 954                         new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
 955                         getContext().dispatcher(), getSelf());
 956
 957                 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
 958
 959             } else if (!shardInformation.isShardInitialized()) {
 960                 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
 961                         shardInformation.getShardName());
 962                 getSender().tell(createNotInitializedException(shardInformation.getShardId()), getSelf());
 963             } else {
 964                 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
 965                         shardInformation.getShardName());
 966                 getSender().tell(new NoShardLeaderException(shardInformation.getShardId()), getSelf());
 967             }
 968
 969             return;
 970         }
 971
 972         getSender().tell(messageSupplier.get(), getSelf());
 973     }
 974
 975     private static NotInitializedException createNotInitializedException(final ShardIdentifier shardId) {
 976         return new NotInitializedException(String.format(
 977                 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
 978     }
 979
 980     @VisibleForTesting
 981     static MemberName memberToName(final Member member) {
 982         return MemberName.forName(member.roles().iterator().next());
 983     }
 984
 985     private void memberRemoved(final ClusterEvent.MemberRemoved message) {
 986         MemberName memberName = memberToName(message.member());
 987
 988         LOG.info("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
 989                 message.member().address());
 990
 991         peerAddressResolver.removePeerAddress(memberName);
 992
 993         for (ShardInformation info : localShards.values()) {
 994             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 995         }
 996     }
 997
 998     private void memberExited(final ClusterEvent.MemberExited message) {
 999         MemberName memberName = memberToName(message.member());
1000
1001         LOG.info("{}: Received MemberExited: memberName: {}, address: {}", persistenceId(), memberName,
1002                 message.member().address());
1003
1004         peerAddressResolver.removePeerAddress(memberName);
1005
1006         for (ShardInformation info : localShards.values()) {
1007             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1008         }
1009     }
1010
1011     private void memberUp(final ClusterEvent.MemberUp message) {
1012         MemberName memberName = memberToName(message.member());
1013
1014         LOG.info("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
1015                 message.member().address());
1016
1017         memberUp(memberName, message.member().address());
1018     }
1019
1020     private void memberUp(final MemberName memberName, final Address address) {
1021         addPeerAddress(memberName, address);
1022         checkReady();
1023     }
1024
1025     private void memberWeaklyUp(final MemberWeaklyUp message) {
1026         MemberName memberName = memberToName(message.member());
1027
1028         LOG.info("{}: Received MemberWeaklyUp: memberName: {}, address: {}", persistenceId(), memberName,
1029                 message.member().address());
1030
1031         memberUp(memberName, message.member().address());
1032     }
1033
1034     private void addPeerAddress(final MemberName memberName, final Address address) {
1035         peerAddressResolver.addPeerAddress(memberName, address);
1036
1037         for (ShardInformation info : localShards.values()) {
1038             String shardName = info.getShardName();
1039             String peerId = getShardIdentifier(memberName, shardName).toString();
1040             info.updatePeerAddress(peerId, peerAddressResolver.getShardActorAddress(shardName, memberName), getSelf());
1041
1042             info.peerUp(memberName, peerId, getSelf());
1043         }
1044     }
1045
1046     private void memberReachable(final ClusterEvent.ReachableMember message) {
1047         MemberName memberName = memberToName(message.member());
1048         LOG.info("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
1049
1050         addPeerAddress(memberName, message.member().address());
1051
1052         markMemberAvailable(memberName);
1053     }
1054
1055     private void memberUnreachable(final ClusterEvent.UnreachableMember message) {
1056         MemberName memberName = memberToName(message.member());
1057         LOG.info("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
1058
1059         markMemberUnavailable(memberName);
1060     }
1061
1062     private void markMemberUnavailable(final MemberName memberName) {
1063         for (ShardInformation info : localShards.values()) {
1064             String leaderId = info.getLeaderId();
1065             if (leaderId != null && ShardIdentifier.fromShardIdString(leaderId).getMemberName().equals(memberName)) {
1066                 LOG.debug("Marking Leader {} as unavailable.", leaderId);
1067                 info.setLeaderAvailable(false);
1068
1069                 primaryShardInfoCache.remove(info.getShardName());
1070
1071                 notifyShardAvailabilityCallbacks(info);
1072             }
1073
1074             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1075         }
1076     }
1077
1078     private void markMemberAvailable(final MemberName memberName) {
1079         for (ShardInformation info : localShards.values()) {
1080             String leaderId = info.getLeaderId();
1081             if (leaderId != null && ShardIdentifier.fromShardIdString(leaderId).getMemberName().equals(memberName)) {
1082                 LOG.debug("Marking Leader {} as available.", leaderId);
1083                 info.setLeaderAvailable(true);
1084             }
1085
1086             info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1087         }
1088     }
1089
1090     private void onDatastoreContextFactory(final DatastoreContextFactory factory) {
1091         datastoreContextFactory = factory;
1092         for (ShardInformation info : localShards.values()) {
1093             info.setDatastoreContext(newShardDatastoreContext(info.getShardName()), getSelf());
1094         }
1095     }
1096
1097     private void onGetLocalShardIds() {
1098         final List<String> response = new ArrayList<>(localShards.size());
1099
1100         for (ShardInformation info : localShards.values()) {
1101             response.add(info.getShardId().toString());
1102         }
1103
1104         getSender().tell(new Status.Success(response), getSelf());
1105     }
1106
1107     private void onSwitchShardBehavior(final SwitchShardBehavior message) {
1108         final ShardIdentifier identifier = message.getShardId();
1109
1110         if (identifier != null) {
1111             final ShardInformation info = localShards.get(identifier.getShardName());
1112             if (info == null) {
1113                 getSender().tell(new Status.Failure(
1114                     new IllegalArgumentException("Shard " + identifier + " is not local")), getSelf());
1115                 return;
1116             }
1117
1118             switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
1119         } else {
1120             for (ShardInformation info : localShards.values()) {
1121                 switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
1122             }
1123         }
1124
1125         getSender().tell(new Status.Success(null), getSelf());
1126     }
1127
1128     private void switchShardBehavior(final ShardInformation info, final SwitchBehavior switchBehavior) {
1129         final ActorRef actor = info.getActor();
1130         if (actor != null) {
1131             actor.tell(switchBehavior, getSelf());
1132         } else {
1133             LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
1134                 info.getShardName(), switchBehavior.getNewState());
1135         }
1136     }
1137
1138     /**
1139      * Notifies all the local shards of a change in the schema context.
1140      *
1141      * @param message the message to send
1142      */
1143     private void updateSchemaContext(final Object message) {
1144         schemaContext = ((UpdateSchemaContext) message).getEffectiveModelContext();
1145
1146         LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getModules().size());
1147
1148         for (ShardInformation info : localShards.values()) {
1149             info.setSchemaContext(schemaContext);
1150
1151             if (info.getActor() == null) {
1152                 LOG.debug("Creating Shard {}", info.getShardId());
1153                 info.setActor(newShardActor(info));
1154                 // Update peer address for every existing peer memeber to avoid missing sending
1155                 // PeerAddressResolved and PeerUp to this shard while UpdateSchemaContext comes after MemberUp.
1156                 String shardName = info.getShardName();
1157                 for (MemberName memberName : peerAddressResolver.getPeerMembers()) {
1158                     String peerId = getShardIdentifier(memberName, shardName).toString() ;
1159                     String peerAddress = peerAddressResolver.getShardActorAddress(shardName, memberName);
1160                     info.updatePeerAddress(peerId, peerAddress, getSelf());
1161                     info.peerUp(memberName, peerId, getSelf());
1162                     LOG.debug("{}: updated peer {} on member {} with address {} on shard {} whose actor address is {}",
1163                             persistenceId(), peerId, memberName, peerAddress, info.getShardId(), info.getActor());
1164                 }
1165             } else {
1166                 info.getActor().tell(message, getSelf());
1167             }
1168         }
1169     }
1170
1171     @VisibleForTesting
1172     protected ClusterWrapper getCluster() {
1173         return cluster;
1174     }
1175
1176     @VisibleForTesting
1177     protected ActorRef newShardActor(final ShardInformation info) {
1178         return getContext().actorOf(info.newProps().withDispatcher(shardDispatcherPath),
1179                 info.getShardId().toString());
1180     }
1181
1182     private void findPrimary(final FindPrimary message) {
1183         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
1184
1185         final String shardName = message.getShardName();
1186         final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
1187
1188         // First see if the there is a local replica for the shard
1189         final ShardInformation info = localShards.get(shardName);
1190         if (info != null && info.isActiveMember()) {
1191             sendResponse(info, message.isWaitUntilReady(), true, () -> {
1192                 String primaryPath = info.getSerializedLeaderActor();
1193                 Object found = canReturnLocalShardState && info.isLeader()
1194                         ? new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
1195                             new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
1196
1197                 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
1198                 return found;
1199             });
1200
1201             return;
1202         }
1203
1204         final Collection<String> visitedAddresses;
1205         if (message instanceof RemoteFindPrimary) {
1206             visitedAddresses = ((RemoteFindPrimary)message).getVisitedAddresses();
1207         } else {
1208             visitedAddresses = new ArrayList<>(1);
1209         }
1210
1211         visitedAddresses.add(peerAddressResolver.getShardManagerActorPathBuilder(cluster.getSelfAddress()).toString());
1212
1213         for (String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
1214             if (visitedAddresses.contains(address)) {
1215                 continue;
1216             }
1217
1218             LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}, visitedAddresses: {}",
1219                     persistenceId(), shardName, address, visitedAddresses);
1220
1221             getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
1222                     message.isWaitUntilReady(), visitedAddresses), getContext());
1223             return;
1224         }
1225
1226         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
1227
1228         getSender().tell(new PrimaryNotFoundException(
1229                 String.format("No primary shard found for %s.", shardName)), getSelf());
1230     }
1231
1232     private void findPrimary(final String shardName, final FindPrimaryResponseHandler handler) {
1233         Timeout findPrimaryTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1234                 .getShardInitializationTimeout().duration().$times(2));
1235
1236         Future<Object> futureObj = ask(getSelf(), new FindPrimary(shardName, true), findPrimaryTimeout);
1237         futureObj.onComplete(new OnComplete<>() {
1238             @Override
1239             public void onComplete(final Throwable failure, final Object response) {
1240                 if (failure != null) {
1241                     handler.onFailure(failure);
1242                 } else {
1243                     if (response instanceof RemotePrimaryShardFound) {
1244                         handler.onRemotePrimaryShardFound((RemotePrimaryShardFound) response);
1245                     } else if (response instanceof LocalPrimaryShardFound) {
1246                         handler.onLocalPrimaryFound((LocalPrimaryShardFound) response);
1247                     } else {
1248                         handler.onUnknownResponse(response);
1249                     }
1250                 }
1251             }
1252         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1253     }
1254
1255     /**
1256      * Construct the name of the shard actor given the name of the member on
1257      * which the shard resides and the name of the shard.
1258      *
1259      * @param memberName the member name
1260      * @param shardName the shard name
1261      * @return a b
1262      */
1263     private ShardIdentifier getShardIdentifier(final MemberName memberName, final String shardName) {
1264         return peerAddressResolver.getShardIdentifier(memberName, shardName);
1265     }
1266
1267     /**
1268      * Create shards that are local to the member on which the ShardManager runs.
1269      */
1270     private void createLocalShards() {
1271         MemberName memberName = this.cluster.getCurrentMemberName();
1272         Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
1273
1274         Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots = new HashMap<>();
1275         if (restoreFromSnapshot != null) {
1276             for (DatastoreSnapshot.ShardSnapshot snapshot: restoreFromSnapshot.getShardSnapshots()) {
1277                 shardSnapshots.put(snapshot.getName(), snapshot);
1278             }
1279         }
1280
1281         // null out to GC
1282         restoreFromSnapshot = null;
1283
1284         for (String shardName : memberShardNames) {
1285             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1286
1287             LOG.debug("{}: Creating local shard: {}", persistenceId(), shardId);
1288
1289             Map<String, String> peerAddresses = getPeerAddresses(shardName);
1290             localShards.put(shardName, createShardInfoFor(shardName, shardId, peerAddresses,
1291                     newShardDatastoreContext(shardName), shardSnapshots));
1292         }
1293     }
1294
1295     @VisibleForTesting
1296     ShardInformation createShardInfoFor(final String shardName, final ShardIdentifier shardId,
1297                                         final Map<String, String> peerAddresses,
1298                                         final DatastoreContext datastoreContext,
1299                                         final Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots) {
1300         return new ShardInformation(shardName, shardId, peerAddresses,
1301                 datastoreContext, Shard.builder().restoreFromSnapshot(shardSnapshots.get(shardName)),
1302                 peerAddressResolver);
1303     }
1304
1305     /**
1306      * Given the name of the shard find the addresses of all it's peers.
1307      *
1308      * @param shardName the shard name
1309      */
1310     Map<String, String> getPeerAddresses(final String shardName) {
1311         final Collection<MemberName> members = configuration.getMembersFromShardName(shardName);
1312         return getPeerAddresses(shardName, members);
1313     }
1314
1315     private Map<String, String> getPeerAddresses(final String shardName, final Collection<MemberName> members) {
1316         Map<String, String> peerAddresses = new HashMap<>();
1317         MemberName currentMemberName = this.cluster.getCurrentMemberName();
1318
1319         for (MemberName memberName : members) {
1320             if (!currentMemberName.equals(memberName)) {
1321                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1322                 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
1323                 peerAddresses.put(shardId.toString(), address);
1324             }
1325         }
1326         return peerAddresses;
1327     }
1328
1329     @Override
1330     public SupervisorStrategy supervisorStrategy() {
1331
1332         return new OneForOneStrategy(10, FiniteDuration.create(1, TimeUnit.MINUTES),
1333                 (Function<Throwable, Directive>) t -> {
1334                     LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
1335                     return SupervisorStrategy.resume();
1336                 });
1337     }
1338
1339     @Override
1340     public String persistenceId() {
1341         return persistenceId;
1342     }
1343
1344     @VisibleForTesting
1345     ShardManagerInfoMBean getMBean() {
1346         return shardManagerMBean;
1347     }
1348
1349     private boolean isShardReplicaOperationInProgress(final String shardName, final ActorRef sender) {
1350         if (shardReplicaOperationsInProgress.contains(shardName)) {
1351             LOG.debug("{}: A shard replica operation for {} is already in progress", persistenceId(), shardName);
1352             sender.tell(new Status.Failure(new IllegalStateException(
1353                 String.format("A shard replica operation for %s is already in progress", shardName))), getSelf());
1354             return true;
1355         }
1356
1357         return false;
1358     }
1359
1360     private void onAddPrefixShardReplica(final AddPrefixShardReplica message) {
1361         LOG.debug("{}: onAddPrefixShardReplica: {}", persistenceId(), message);
1362
1363         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1364                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1365         final String shardName = shardId.getShardName();
1366
1367         // Create the localShard
1368         if (schemaContext == null) {
1369             LOG.debug("{}: No SchemaContext is available in order to create a local shard instance for {}",
1370                 persistenceId(), shardName);
1371             getSender().tell(new Status.Failure(new IllegalStateException(
1372                 "No SchemaContext is available in order to create a local shard instance for " + shardName)),
1373                 getSelf());
1374             return;
1375         }
1376
1377         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1378                 getSelf()) {
1379             @Override
1380             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1381                 final RunnableMessage runnable = (RunnableMessage) () -> addPrefixShard(getShardName(),
1382                         message.getShardPrefix(), response, getSender());
1383                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1384                     getSelf().tell(runnable, getTargetActor());
1385                 }
1386             }
1387
1388             @Override
1389             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1390                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1391             }
1392         });
1393     }
1394
1395     private void onAddShardReplica(final AddShardReplica shardReplicaMsg) {
1396         final String shardName = shardReplicaMsg.getShardName();
1397
1398         LOG.debug("{}: onAddShardReplica: {}", persistenceId(), shardReplicaMsg);
1399
1400         // verify the shard with the specified name is present in the cluster configuration
1401         if (!this.configuration.isShardConfigured(shardName)) {
1402             LOG.debug("{}: No module configuration exists for shard {}", persistenceId(), shardName);
1403             getSender().tell(new Status.Failure(new IllegalArgumentException(
1404                 "No module configuration exists for shard " + shardName)), getSelf());
1405             return;
1406         }
1407
1408         // Create the localShard
1409         if (schemaContext == null) {
1410             LOG.debug("{}: No SchemaContext is available in order to create a local shard instance for {}",
1411                 persistenceId(), shardName);
1412             getSender().tell(new Status.Failure(new IllegalStateException(
1413                 "No SchemaContext is available in order to create a local shard instance for " + shardName)),
1414                 getSelf());
1415             return;
1416         }
1417
1418         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1419                 getSelf()) {
1420             @Override
1421             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1422                 final RunnableMessage runnable = (RunnableMessage) () ->
1423                     addShard(getShardName(), response, getSender());
1424                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1425                     getSelf().tell(runnable, getTargetActor());
1426                 }
1427             }
1428
1429             @Override
1430             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1431                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1432             }
1433         });
1434     }
1435
1436     @SuppressFBWarnings(value = "UPM_UNCALLED_PRIVATE_METHOD",
1437             justification = "https://github.com/spotbugs/spotbugs/issues/811")
1438     private void sendLocalReplicaAlreadyExistsReply(final String shardName, final ActorRef sender) {
1439         LOG.debug("{}: Local shard {} already exists", persistenceId(), shardName);
1440         sender.tell(new Status.Failure(new AlreadyExistsException(
1441             String.format("Local shard %s already exists", shardName))), getSelf());
1442     }
1443
1444     @SuppressFBWarnings(value = "UPM_UNCALLED_PRIVATE_METHOD",
1445             justification = "https://github.com/spotbugs/spotbugs/issues/811")
1446     private void addPrefixShard(final String shardName, final YangInstanceIdentifier shardPrefix,
1447                                 final RemotePrimaryShardFound response, final ActorRef sender) {
1448         if (isShardReplicaOperationInProgress(shardName, sender)) {
1449             return;
1450         }
1451
1452         shardReplicaOperationsInProgress.add(shardName);
1453
1454         final ShardInformation shardInfo;
1455         final boolean removeShardOnFailure;
1456         ShardInformation existingShardInfo = localShards.get(shardName);
1457         if (existingShardInfo == null) {
1458             removeShardOnFailure = true;
1459             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1460
1461             final Builder builder = newShardDatastoreContextBuilder(shardName);
1462             builder.storeRoot(shardPrefix).customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName());
1463
1464             DatastoreContext datastoreContext = builder.build();
1465
1466             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1467                     Shard.builder(), peerAddressResolver);
1468             shardInfo.setActiveMember(false);
1469             shardInfo.setSchemaContext(schemaContext);
1470             localShards.put(shardName, shardInfo);
1471             shardInfo.setActor(newShardActor(shardInfo));
1472         } else {
1473             removeShardOnFailure = false;
1474             shardInfo = existingShardInfo;
1475         }
1476
1477         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1478     }
1479
1480     @SuppressFBWarnings(value = "UPM_UNCALLED_PRIVATE_METHOD",
1481             justification = "https://github.com/spotbugs/spotbugs/issues/811")
1482     private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) {
1483         if (isShardReplicaOperationInProgress(shardName, sender)) {
1484             return;
1485         }
1486
1487         shardReplicaOperationsInProgress.add(shardName);
1488
1489         final ShardInformation shardInfo;
1490         final boolean removeShardOnFailure;
1491         ShardInformation existingShardInfo = localShards.get(shardName);
1492         if (existingShardInfo == null) {
1493             removeShardOnFailure = true;
1494             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1495
1496             DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName)
1497                     .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
1498
1499             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1500                     Shard.builder(), peerAddressResolver);
1501             shardInfo.setActiveMember(false);
1502             shardInfo.setSchemaContext(schemaContext);
1503             localShards.put(shardName, shardInfo);
1504             shardInfo.setActor(newShardActor(shardInfo));
1505         } else {
1506             removeShardOnFailure = false;
1507             shardInfo = existingShardInfo;
1508         }
1509
1510         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1511     }
1512
1513     private void execAddShard(final String shardName,
1514                               final ShardInformation shardInfo,
1515                               final RemotePrimaryShardFound response,
1516                               final boolean removeShardOnFailure,
1517                               final ActorRef sender) {
1518
1519         final String localShardAddress =
1520                 peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName());
1521
1522         //inform ShardLeader to add this shard as a replica by sending an AddServer message
1523         LOG.debug("{}: Sending AddServer message to peer {} for shard {}", persistenceId(),
1524                 response.getPrimaryPath(), shardInfo.getShardId());
1525
1526         final Timeout addServerTimeout = new Timeout(shardInfo.getDatastoreContext()
1527                 .getShardLeaderElectionTimeout().duration());
1528         final Future<Object> futureObj = ask(getContext().actorSelection(response.getPrimaryPath()),
1529                 new AddServer(shardInfo.getShardId().toString(), localShardAddress, true), addServerTimeout);
1530
1531         futureObj.onComplete(new OnComplete<>() {
1532             @Override
1533             public void onComplete(final Throwable failure, final Object addServerResponse) {
1534                 if (failure != null) {
1535                     LOG.debug("{}: AddServer request to {} for {} failed", persistenceId(),
1536                             response.getPrimaryPath(), shardName, failure);
1537
1538                     final String msg = String.format("AddServer request to leader %s for shard %s failed",
1539                             response.getPrimaryPath(), shardName);
1540                     self().tell(new ForwardedAddServerFailure(shardName, msg, failure, removeShardOnFailure), sender);
1541                 } else {
1542                     self().tell(new ForwardedAddServerReply(shardInfo, (AddServerReply)addServerResponse,
1543                             response.getPrimaryPath(), removeShardOnFailure), sender);
1544                 }
1545             }
1546         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1547     }
1548
1549     private void onAddServerFailure(final String shardName, final String message, final Throwable failure,
1550             final ActorRef sender, final boolean removeShardOnFailure) {
1551         shardReplicaOperationsInProgress.remove(shardName);
1552
1553         if (removeShardOnFailure) {
1554             ShardInformation shardInfo = localShards.remove(shardName);
1555             if (shardInfo.getActor() != null) {
1556                 shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf());
1557             }
1558         }
1559
1560         sender.tell(new Status.Failure(message == null ? failure :
1561             new RuntimeException(message, failure)), getSelf());
1562     }
1563
1564     private void onAddServerReply(final ShardInformation shardInfo, final AddServerReply replyMsg,
1565             final ActorRef sender, final String leaderPath, final boolean removeShardOnFailure) {
1566         String shardName = shardInfo.getShardName();
1567         shardReplicaOperationsInProgress.remove(shardName);
1568
1569         LOG.debug("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath);
1570
1571         if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1572             LOG.debug("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName);
1573
1574             // Make the local shard voting capable
1575             shardInfo.setDatastoreContext(newShardDatastoreContext(shardName), getSelf());
1576             shardInfo.setActiveMember(true);
1577             persistShardList();
1578
1579             sender.tell(new Status.Success(null), getSelf());
1580         } else if (replyMsg.getStatus() == ServerChangeStatus.ALREADY_EXISTS) {
1581             sendLocalReplicaAlreadyExistsReply(shardName, sender);
1582         } else {
1583             LOG.warn("{}: Leader failed to add shard replica {} with status {}",
1584                     persistenceId(), shardName, replyMsg.getStatus());
1585
1586             Exception failure = getServerChangeException(AddServer.class, replyMsg.getStatus(), leaderPath,
1587                     shardInfo.getShardId());
1588
1589             onAddServerFailure(shardName, null, failure, sender, removeShardOnFailure);
1590         }
1591     }
1592
1593     private static Exception getServerChangeException(final Class<?> serverChange,
1594             final ServerChangeStatus serverChangeStatus, final String leaderPath, final ShardIdentifier shardId) {
1595         switch (serverChangeStatus) {
1596             case TIMEOUT:
1597                 return new TimeoutException(String.format(
1598                         "The shard leader %s timed out trying to replicate the initial data to the new shard %s."
1599                         + "Possible causes - there was a problem replicating the data or shard leadership changed "
1600                         + "while replicating the shard data", leaderPath, shardId.getShardName()));
1601             case NO_LEADER:
1602                 return new NoShardLeaderException(shardId);
1603             case NOT_SUPPORTED:
1604                 return new UnsupportedOperationException(String.format("%s request is not supported for shard %s",
1605                         serverChange.getSimpleName(), shardId.getShardName()));
1606             default :
1607                 return new RuntimeException(String.format("%s request to leader %s for shard %s failed with status %s",
1608                         serverChange.getSimpleName(), leaderPath, shardId.getShardName(), serverChangeStatus));
1609         }
1610     }
1611
1612     private void onRemoveShardReplica(final RemoveShardReplica shardReplicaMsg) {
1613         LOG.debug("{}: onRemoveShardReplica: {}", persistenceId(), shardReplicaMsg);
1614
1615         findPrimary(shardReplicaMsg.getShardName(), new AutoFindPrimaryFailureResponseHandler(getSender(),
1616                 shardReplicaMsg.getShardName(), persistenceId(), getSelf()) {
1617             @Override
1618             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1619                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1620             }
1621
1622             @Override
1623             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1624                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1625             }
1626
1627             private void doRemoveShardReplicaAsync(final String primaryPath) {
1628                 getSelf().tell((RunnableMessage) () -> removeShardReplica(shardReplicaMsg, getShardName(),
1629                         primaryPath, getSender()), getTargetActor());
1630             }
1631         });
1632     }
1633
1634     private void onRemovePrefixShardReplica(final RemovePrefixShardReplica message) {
1635         LOG.debug("{}: onRemovePrefixShardReplica: {}", persistenceId(), message);
1636
1637         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1638                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1639         final String shardName = shardId.getShardName();
1640
1641         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(),
1642                 shardName, persistenceId(), getSelf()) {
1643             @Override
1644             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1645                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1646             }
1647
1648             @Override
1649             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1650                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1651             }
1652
1653             private void doRemoveShardReplicaAsync(final String primaryPath) {
1654                 getSelf().tell((RunnableMessage) () -> removePrefixShardReplica(message, getShardName(),
1655                         primaryPath, getSender()), getTargetActor());
1656             }
1657         });
1658     }
1659
1660     private void persistShardList() {
1661         List<String> shardList = new ArrayList<>(localShards.keySet());
1662         for (ShardInformation shardInfo : localShards.values()) {
1663             if (!shardInfo.isActiveMember()) {
1664                 shardList.remove(shardInfo.getShardName());
1665             }
1666         }
1667         LOG.debug("{}: persisting the shard list {}", persistenceId(), shardList);
1668         saveSnapshot(updateShardManagerSnapshot(shardList, configuration.getAllPrefixShardConfigurations()));
1669     }
1670
1671     private ShardManagerSnapshot updateShardManagerSnapshot(
1672             final List<String> shardList,
1673             final Map<DOMDataTreeIdentifier, PrefixShardConfiguration> allPrefixShardConfigurations) {
1674         currentSnapshot = new ShardManagerSnapshot(shardList, allPrefixShardConfigurations);
1675         return currentSnapshot;
1676     }
1677
1678     private void applyShardManagerSnapshot(final ShardManagerSnapshot snapshot) {
1679         currentSnapshot = snapshot;
1680
1681         LOG.debug("{}: onSnapshotOffer: {}", persistenceId(), currentSnapshot);
1682
1683         final MemberName currentMember = cluster.getCurrentMemberName();
1684         Set<String> configuredShardList =
1685             new HashSet<>(configuration.getMemberShardNames(currentMember));
1686         for (String shard : currentSnapshot.getShardList()) {
1687             if (!configuredShardList.contains(shard)) {
1688                 // add the current member as a replica for the shard
1689                 LOG.debug("{}: adding shard {}", persistenceId(), shard);
1690                 configuration.addMemberReplicaForShard(shard, currentMember);
1691             } else {
1692                 configuredShardList.remove(shard);
1693             }
1694         }
1695         for (String shard : configuredShardList) {
1696             // remove the member as a replica for the shard
1697             LOG.debug("{}: removing shard {}", persistenceId(), shard);
1698             configuration.removeMemberReplicaForShard(shard, currentMember);
1699         }
1700     }
1701
1702     private void onSaveSnapshotSuccess(final SaveSnapshotSuccess successMessage) {
1703         LOG.debug("{} saved ShardManager snapshot successfully. Deleting the prev snapshot if available",
1704             persistenceId());
1705         deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), successMessage.metadata().timestamp() - 1,
1706             0, 0));
1707     }
1708
1709     private void onChangeShardServersVotingStatus(final ChangeShardMembersVotingStatus changeMembersVotingStatus) {
1710         LOG.debug("{}: onChangeShardServersVotingStatus: {}", persistenceId(), changeMembersVotingStatus);
1711
1712         String shardName = changeMembersVotingStatus.getShardName();
1713         Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1714         for (Entry<String, Boolean> e: changeMembersVotingStatus.getMeberVotingStatusMap().entrySet()) {
1715             serverVotingStatusMap.put(getShardIdentifier(MemberName.forName(e.getKey()), shardName).toString(),
1716                     e.getValue());
1717         }
1718
1719         ChangeServersVotingStatus changeServersVotingStatus = new ChangeServersVotingStatus(serverVotingStatusMap);
1720
1721         findLocalShard(shardName, getSender(),
1722             localShardFound -> changeShardMembersVotingStatus(changeServersVotingStatus, shardName,
1723             localShardFound.getPath(), getSender()));
1724     }
1725
1726     private void onFlipShardMembersVotingStatus(final FlipShardMembersVotingStatus flipMembersVotingStatus) {
1727         LOG.debug("{}: onFlipShardMembersVotingStatus: {}", persistenceId(), flipMembersVotingStatus);
1728
1729         ActorRef sender = getSender();
1730         final String shardName = flipMembersVotingStatus.getShardName();
1731         findLocalShard(shardName, sender, localShardFound -> {
1732             Future<Object> future = ask(localShardFound.getPath(), GetOnDemandRaftState.INSTANCE,
1733                     Timeout.apply(30, TimeUnit.SECONDS));
1734
1735             future.onComplete(new OnComplete<>() {
1736                 @Override
1737                 public void onComplete(final Throwable failure, final Object response) {
1738                     if (failure != null) {
1739                         sender.tell(new Status.Failure(new RuntimeException(
1740                                 String.format("Failed to access local shard %s", shardName), failure)), self());
1741                         return;
1742                     }
1743
1744                     OnDemandRaftState raftState = (OnDemandRaftState) response;
1745                     Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1746                     for (Entry<String, Boolean> e: raftState.getPeerVotingStates().entrySet()) {
1747                         serverVotingStatusMap.put(e.getKey(), !e.getValue());
1748                     }
1749
1750                     serverVotingStatusMap.put(getShardIdentifier(cluster.getCurrentMemberName(), shardName)
1751                             .toString(), !raftState.isVoting());
1752
1753                     changeShardMembersVotingStatus(new ChangeServersVotingStatus(serverVotingStatusMap),
1754                             shardName, localShardFound.getPath(), sender);
1755                 }
1756             }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1757         });
1758
1759     }
1760
1761     private void findLocalShard(final FindLocalShard message) {
1762         LOG.debug("{}: findLocalShard : {}", persistenceId(), message.getShardName());
1763
1764         final ShardInformation shardInformation = localShards.get(message.getShardName());
1765
1766         if (shardInformation == null) {
1767             LOG.debug("{}: Local shard {} not found - shards present: {}",
1768                     persistenceId(), message.getShardName(), localShards.keySet());
1769
1770             getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
1771             return;
1772         }
1773
1774         sendResponse(shardInformation, message.isWaitUntilInitialized(), false,
1775             () -> new LocalShardFound(shardInformation.getActor()));
1776     }
1777
1778     private void findLocalShard(final String shardName, final ActorRef sender,
1779             final Consumer<LocalShardFound> onLocalShardFound) {
1780         Timeout findLocalTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1781                 .getShardInitializationTimeout().duration().$times(2));
1782
1783         Future<Object> futureObj = ask(getSelf(), new FindLocalShard(shardName, true), findLocalTimeout);
1784         futureObj.onComplete(new OnComplete<>() {
1785             @Override
1786             public void onComplete(final Throwable failure, final Object response) {
1787                 if (failure != null) {
1788                     LOG.debug("{}: Received failure from FindLocalShard for shard {}", persistenceId, shardName,
1789                             failure);
1790                     sender.tell(new Status.Failure(new RuntimeException(
1791                             String.format("Failed to find local shard %s", shardName), failure)), self());
1792                 } else {
1793                     if (response instanceof LocalShardFound) {
1794                         getSelf().tell((RunnableMessage) () -> onLocalShardFound.accept((LocalShardFound) response),
1795                                 sender);
1796                     } else if (response instanceof LocalShardNotFound) {
1797                         LOG.debug("{}: Local shard {} does not exist", persistenceId, shardName);
1798                         sender.tell(new Status.Failure(new IllegalArgumentException(
1799                             String.format("Local shard %s does not exist", shardName))), self());
1800                     } else {
1801                         LOG.debug("{}: Failed to find local shard {}: received response: {}", persistenceId, shardName,
1802                             response);
1803                         sender.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response
1804                                 : new RuntimeException(
1805                                     String.format("Failed to find local shard %s: received response: %s", shardName,
1806                                         response))), self());
1807                     }
1808                 }
1809             }
1810         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1811     }
1812
1813     private void changeShardMembersVotingStatus(final ChangeServersVotingStatus changeServersVotingStatus,
1814             final String shardName, final ActorRef shardActorRef, final ActorRef sender) {
1815         if (isShardReplicaOperationInProgress(shardName, sender)) {
1816             return;
1817         }
1818
1819         shardReplicaOperationsInProgress.add(shardName);
1820
1821         DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
1822         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1823
1824         LOG.debug("{}: Sending ChangeServersVotingStatus message {} to local shard {}", persistenceId(),
1825                 changeServersVotingStatus, shardActorRef.path());
1826
1827         Timeout timeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().duration().$times(2));
1828         Future<Object> futureObj = ask(shardActorRef, changeServersVotingStatus, timeout);
1829
1830         futureObj.onComplete(new OnComplete<>() {
1831             @Override
1832             public void onComplete(final Throwable failure, final Object response) {
1833                 shardReplicaOperationsInProgress.remove(shardName);
1834                 if (failure != null) {
1835                     LOG.debug("{}: ChangeServersVotingStatus request to local shard {} failed", persistenceId(),
1836                         shardActorRef.path(), failure);
1837                     sender.tell(new Status.Failure(new RuntimeException(
1838                         String.format("ChangeServersVotingStatus request to local shard %s failed",
1839                             shardActorRef.path()), failure)), self());
1840                 } else {
1841                     LOG.debug("{}: Received {} from local shard {}", persistenceId(), response, shardActorRef.path());
1842
1843                     ServerChangeReply replyMsg = (ServerChangeReply) response;
1844                     if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1845                         LOG.debug("{}: ChangeServersVotingStatus succeeded for shard {}", persistenceId(), shardName);
1846                         sender.tell(new Status.Success(null), getSelf());
1847                     } else if (replyMsg.getStatus() == ServerChangeStatus.INVALID_REQUEST) {
1848                         sender.tell(new Status.Failure(new IllegalArgumentException(String.format(
1849                                 "The requested voting state change for shard %s is invalid. At least one member "
1850                                 + "must be voting", shardId.getShardName()))), getSelf());
1851                     } else {
1852                         LOG.warn("{}: ChangeServersVotingStatus failed for shard {} with status {}",
1853                                 persistenceId(), shardName, replyMsg.getStatus());
1854
1855                         Exception error = getServerChangeException(ChangeServersVotingStatus.class,
1856                                 replyMsg.getStatus(), shardActorRef.path().toString(), shardId);
1857                         sender.tell(new Status.Failure(error), getSelf());
1858                     }
1859                 }
1860             }
1861         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1862     }
1863
1864     private static final class ForwardedAddServerReply {
1865         ShardInformation shardInfo;
1866         AddServerReply addServerReply;
1867         String leaderPath;
1868         boolean removeShardOnFailure;
1869
1870         ForwardedAddServerReply(final ShardInformation shardInfo, final AddServerReply addServerReply,
1871             final String leaderPath, final boolean removeShardOnFailure) {
1872             this.shardInfo = shardInfo;
1873             this.addServerReply = addServerReply;
1874             this.leaderPath = leaderPath;
1875             this.removeShardOnFailure = removeShardOnFailure;
1876         }
1877     }
1878
1879     private static final class ForwardedAddServerFailure {
1880         String shardName;
1881         String failureMessage;
1882         Throwable failure;
1883         boolean removeShardOnFailure;
1884
1885         ForwardedAddServerFailure(final String shardName, final String failureMessage, final Throwable failure,
1886                 final boolean removeShardOnFailure) {
1887             this.shardName = shardName;
1888             this.failureMessage = failureMessage;
1889             this.failure = failure;
1890             this.removeShardOnFailure = removeShardOnFailure;
1891         }
1892     }
1893
1894     static class OnShardInitialized {
1895         private final Runnable replyRunnable;
1896         private Cancellable timeoutSchedule;
1897
1898         OnShardInitialized(final Runnable replyRunnable) {
1899             this.replyRunnable = replyRunnable;
1900         }
1901
1902         Runnable getReplyRunnable() {
1903             return replyRunnable;
1904         }
1905
1906         Cancellable getTimeoutSchedule() {
1907             return timeoutSchedule;
1908         }
1909
1910         void setTimeoutSchedule(final Cancellable timeoutSchedule) {
1911             this.timeoutSchedule = timeoutSchedule;
1912         }
1913     }
1914
1915     static class OnShardReady extends OnShardInitialized {
1916         OnShardReady(final Runnable replyRunnable) {
1917             super(replyRunnable);
1918         }
1919     }
1920
1921     private interface RunnableMessage extends Runnable {
1922     }
1923
1924     /**
1925      * The FindPrimaryResponseHandler provides specific callback methods which are invoked when a response to the
1926      * a remote or local find primary message is processed.
1927      */
1928     private interface FindPrimaryResponseHandler {
1929         /**
1930          * Invoked when a Failure message is received as a response.
1931          *
1932          * @param failure the failure exception
1933          */
1934         void onFailure(Throwable failure);
1935
1936         /**
1937          * Invoked when a RemotePrimaryShardFound response is received.
1938          *
1939          * @param response the response
1940          */
1941         void onRemotePrimaryShardFound(RemotePrimaryShardFound response);
1942
1943         /**
1944          * Invoked when a LocalPrimaryShardFound response is received.
1945          *
1946          * @param response the response
1947          */
1948         void onLocalPrimaryFound(LocalPrimaryShardFound response);
1949
1950         /**
1951          * Invoked when an unknown response is received. This is another type of failure.
1952          *
1953          * @param response the response
1954          */
1955         void onUnknownResponse(Object response);
1956     }
1957
1958     /**
1959      * The AutoFindPrimaryFailureResponseHandler automatically processes Failure responses when finding a primary
1960      * replica and sends a wrapped Failure response to some targetActor.
1961      */
1962     private abstract static class AutoFindPrimaryFailureResponseHandler implements FindPrimaryResponseHandler {
1963         private final ActorRef targetActor;
1964         private final String shardName;
1965         private final String persistenceId;
1966         private final ActorRef shardManagerActor;
1967
1968         /**
1969          * Constructs an instance.
1970          *
1971          * @param targetActor The actor to whom the Failure response should be sent when a FindPrimary failure occurs
1972          * @param shardName The name of the shard for which the primary replica had to be found
1973          * @param persistenceId The persistenceId for the ShardManager
1974          * @param shardManagerActor The ShardManager actor which triggered the call to FindPrimary
1975          */
1976         protected AutoFindPrimaryFailureResponseHandler(final ActorRef targetActor, final String shardName,
1977                 final String persistenceId, final ActorRef shardManagerActor) {
1978             this.targetActor = requireNonNull(targetActor);
1979             this.shardName = requireNonNull(shardName);
1980             this.persistenceId = requireNonNull(persistenceId);
1981             this.shardManagerActor = requireNonNull(shardManagerActor);
1982         }
1983
1984         public ActorRef getTargetActor() {
1985             return targetActor;
1986         }
1987
1988         public String getShardName() {
1989             return shardName;
1990         }
1991
1992         @Override
1993         public void onFailure(final Throwable failure) {
1994             LOG.debug("{}: Received failure from FindPrimary for shard {}", persistenceId, shardName, failure);
1995             targetActor.tell(new Status.Failure(new RuntimeException(
1996                     String.format("Failed to find leader for shard %s", shardName), failure)), shardManagerActor);
1997         }
1998
1999         @Override
2000         public void onUnknownResponse(final Object response) {
2001             LOG.debug("{}: Failed to find leader for shard {}: received response: {}", persistenceId, shardName,
2002                 response);
2003             targetActor.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response
2004                     : new RuntimeException(String.format("Failed to find leader for shard %s: received response: %s",
2005                         shardName, response))), shardManagerActor);
2006         }
2007     }
2008
2009     /**
2010      * The WrappedShardResponse class wraps a response from a Shard.
2011      */
2012     private static final class WrappedShardResponse {
2013         private final ShardIdentifier shardId;
2014         private final Object response;
2015         private final String leaderPath;
2016
2017         WrappedShardResponse(final ShardIdentifier shardId, final Object response, final String leaderPath) {
2018             this.shardId = shardId;
2019             this.response = response;
2020             this.leaderPath = leaderPath;
2021         }
2022
2023         ShardIdentifier getShardId() {
2024             return shardId;
2025         }
2026
2027         Object getResponse() {
2028             return response;
2029         }
2030
2031         String getLeaderPath() {
2032             return leaderPath;
2033         }
2034     }
2035
2036     private static final class ShardNotInitializedTimeout {
2037         private final ActorRef sender;
2038         private final ShardInformation shardInfo;
2039         private final OnShardInitialized onShardInitialized;
2040
2041         ShardNotInitializedTimeout(final ShardInformation shardInfo, final OnShardInitialized onShardInitialized,
2042             final ActorRef sender) {
2043             this.sender = sender;
2044             this.shardInfo = shardInfo;
2045             this.onShardInitialized = onShardInitialized;
2046         }
2047
2048         ActorRef getSender() {
2049             return sender;
2050         }
2051
2052         ShardInformation getShardInfo() {
2053             return shardInfo;
2054         }
2055
2056         OnShardInitialized getOnShardInitialized() {
2057             return onShardInitialized;
2058         }
2059     }
2060 }
2061
2062
2063