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