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 datastoreType =
 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, datastoreType);
 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                         // At any rate, invalidate primaryShardInfo cache
 535                         primaryShardInfoCache.remove(shardName);
 536
 537                         shardActorsStopping.remove(shardName);
 538                         notifyOnCompleteTasks(failure, result);
 539                     }, ActorRef.noSender());
 540                 }
 541             };
 542
 543             shardActorsStopping.put(shardName, onComplete);
 544             stopFuture.onComplete(onComplete, new Dispatchers(context().system().dispatchers())
 545                     .getDispatcher(Dispatchers.DispatcherType.Client));
 546         }
 547
 548         LOG.debug("{} : Local Shard replica for shard {} has been removed", persistenceId(), shardName);
 549         persistShardList();
 550     }
 551
 552     private void onGetSnapshot() {
 553         LOG.debug("{}: onGetSnapshot", persistenceId());
 554
 555         List<String> notInitialized = null;
 556         for (ShardInformation shardInfo : localShards.values()) {
 557             if (!shardInfo.isShardInitialized()) {
 558                 if (notInitialized == null) {
 559                     notInitialized = new ArrayList<>();
 560                 }
 561
 562                 notInitialized.add(shardInfo.getShardName());
 563             }
 564         }
 565
 566         if (notInitialized != null) {
 567             getSender().tell(new Status.Failure(new IllegalStateException(String.format(
 568                     "%d shard(s) %s are not initialized", notInitialized.size(), notInitialized))), getSelf());
 569             return;
 570         }
 571
 572         ActorRef replyActor = getContext().actorOf(ShardManagerGetSnapshotReplyActor.props(
 573                 new ArrayList<>(localShards.keySet()), type, currentSnapshot , getSender(), persistenceId(),
 574                 datastoreContextFactory.getBaseDatastoreContext().getShardInitializationTimeout().duration()));
 575
 576         for (ShardInformation shardInfo: localShards.values()) {
 577             shardInfo.getActor().tell(GetSnapshot.INSTANCE, replyActor);
 578         }
 579     }
 580
 581     @SuppressWarnings("checkstyle:IllegalCatch")
 582     private void onCreateShard(final CreateShard createShard) {
 583         LOG.debug("{}: onCreateShard: {}", persistenceId(), createShard);
 584
 585         Object reply;
 586         try {
 587             String shardName = createShard.getModuleShardConfig().getShardName();
 588             if (localShards.containsKey(shardName)) {
 589                 LOG.debug("{}: Shard {} already exists", persistenceId(), shardName);
 590                 reply = new Status.Success(String.format("Shard with name %s already exists", shardName));
 591             } else {
 592                 doCreateShard(createShard);
 593                 reply = new Status.Success(null);
 594             }
 595         } catch (Exception e) {
 596             LOG.error("{}: onCreateShard failed", persistenceId(), e);
 597             reply = new Status.Failure(e);
 598         }
 599
 600         if (getSender() != null && !getContext().system().deadLetters().equals(getSender())) {
 601             getSender().tell(reply, getSelf());
 602         }
 603     }
 604
 605     private void onPrefixShardCreated(final PrefixShardCreated message) {
 606         LOG.debug("{}: onPrefixShardCreated: {}", persistenceId(), message);
 607
 608         final PrefixShardConfiguration config = message.getConfiguration();
 609         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
 610                 ClusterUtils.getCleanShardName(config.getPrefix().getRootIdentifier()));
 611         final String shardName = shardId.getShardName();
 612
 613         if (isPreviousShardActorStopInProgress(shardName, message)) {
 614             return;
 615         }
 616
 617         if (localShards.containsKey(shardName)) {
 618             LOG.debug("{}: Received create for an already existing shard {}", persistenceId(), shardName);
 619             final PrefixShardConfiguration existing =
 620                     configuration.getAllPrefixShardConfigurations().get(config.getPrefix());
 621
 622             if (existing != null && existing.equals(config)) {
 623                 // we don't have to do nothing here
 624                 return;
 625             }
 626         }
 627
 628         doCreatePrefixShard(config, shardId, shardName);
 629     }
 630
 631     private boolean isPreviousShardActorStopInProgress(final String shardName, final Object messageToDefer) {
 632         final CompositeOnComplete<Boolean> stopOnComplete = shardActorsStopping.get(shardName);
 633         if (stopOnComplete == null) {
 634             return false;
 635         }
 636
 637         LOG.debug("{} : Stop is in progress for shard {} - adding OnComplete callback to defer {}", persistenceId(),
 638                 shardName, messageToDefer);
 639         final ActorRef sender = getSender();
 640         stopOnComplete.addOnComplete(new OnComplete<Boolean>() {
 641             @Override
 642             public void onComplete(final Throwable failure, final Boolean result) {
 643                 LOG.debug("{} : Stop complete for shard {} - re-queing {}", persistenceId(), shardName, messageToDefer);
 644                 self().tell(messageToDefer, sender);
 645             }
 646         });
 647
 648         return true;
 649     }
 650
 651     private void doCreatePrefixShard(final PrefixShardConfiguration config, final ShardIdentifier shardId,
 652             final String shardName) {
 653         configuration.addPrefixShardConfiguration(config);
 654
 655         final Builder builder = newShardDatastoreContextBuilder(shardName);
 656         builder.logicalStoreType(LogicalDatastoreType.valueOf(config.getPrefix().getDatastoreType().name()))
 657                 .storeRoot(config.getPrefix().getRootIdentifier());
 658         DatastoreContext shardDatastoreContext = builder.build();
 659
 660         final Map<String, String> peerAddresses = getPeerAddresses(shardName);
 661         final boolean isActiveMember = true;
 662
 663         LOG.debug("{} doCreatePrefixShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
 664                 persistenceId(), shardId, config.getShardMemberNames(), peerAddresses, isActiveMember);
 665
 666         final ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
 667                 shardDatastoreContext, Shard.builder(), peerAddressResolver);
 668         info.setActiveMember(isActiveMember);
 669         localShards.put(info.getShardName(), info);
 670
 671         if (schemaContext != null) {
 672             info.setSchemaContext(schemaContext);
 673             info.setActor(newShardActor(info));
 674         }
 675     }
 676
 677     private void onPrefixShardRemoved(final PrefixShardRemoved message) {
 678         LOG.debug("{}: onPrefixShardRemoved : {}", persistenceId(), message);
 679
 680         final DOMDataTreeIdentifier prefix = message.getPrefix();
 681         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
 682                 ClusterUtils.getCleanShardName(prefix.getRootIdentifier()));
 683
 684         configuration.removePrefixShardConfiguration(prefix);
 685         removeShard(shardId);
 686     }
 687
 688     private void doCreateShard(final CreateShard createShard) {
 689         final ModuleShardConfiguration moduleShardConfig = createShard.getModuleShardConfig();
 690         final String shardName = moduleShardConfig.getShardName();
 691
 692         configuration.addModuleShardConfiguration(moduleShardConfig);
 693
 694         DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
 695         if (shardDatastoreContext == null) {
 696             shardDatastoreContext = newShardDatastoreContext(shardName);
 697         } else {
 698             shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).shardPeerAddressResolver(
 699                     peerAddressResolver).build();
 700         }
 701
 702         ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
 703
 704         boolean shardWasInRecoveredSnapshot = currentSnapshot != null
 705                 && currentSnapshot.getShardList().contains(shardName);
 706
 707         Map<String, String> peerAddresses;
 708         boolean isActiveMember;
 709         if (shardWasInRecoveredSnapshot || configuration.getMembersFromShardName(shardName)
 710                 .contains(cluster.getCurrentMemberName())) {
 711             peerAddresses = getPeerAddresses(shardName);
 712             isActiveMember = true;
 713         } else {
 714             // The local member is not in the static shard member configuration and the shard did not
 715             // previously exist (ie !shardWasInRecoveredSnapshot). In this case we'll create
 716             // the shard with no peers and with elections disabled so it stays as follower. A
 717             // subsequent AddServer request will be needed to make it an active member.
 718             isActiveMember = false;
 719             peerAddresses = Collections.emptyMap();
 720             shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext)
 721                     .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
 722         }
 723
 724         LOG.debug("{} doCreateShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
 725                 persistenceId(), shardId, moduleShardConfig.getShardMemberNames(), peerAddresses,
 726                 isActiveMember);
 727
 728         ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
 729                 shardDatastoreContext, createShard.getShardBuilder(), peerAddressResolver);
 730         info.setActiveMember(isActiveMember);
 731         localShards.put(info.getShardName(), info);
 732
 733         if (schemaContext != null) {
 734             info.setSchemaContext(schemaContext);
 735             info.setActor(newShardActor(info));
 736         }
 737     }
 738
 739     private DatastoreContext.Builder newShardDatastoreContextBuilder(final String shardName) {
 740         return DatastoreContext.newBuilderFrom(datastoreContextFactory.getShardDatastoreContext(shardName))
 741                 .shardPeerAddressResolver(peerAddressResolver);
 742     }
 743
 744     private DatastoreContext newShardDatastoreContext(final String shardName) {
 745         return newShardDatastoreContextBuilder(shardName).build();
 746     }
 747
 748     private void checkReady() {
 749         if (isReadyWithLeaderId()) {
 750             LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
 751                     persistenceId(), type, waitTillReadyCountdownLatch.getCount());
 752
 753             waitTillReadyCountdownLatch.countDown();
 754         }
 755     }
 756
 757     private void onLeaderStateChanged(final ShardLeaderStateChanged leaderStateChanged) {
 758         LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
 759
 760         ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
 761         if (shardInformation != null) {
 762             shardInformation.setLocalDataTree(leaderStateChanged.getLocalShardDataTree());
 763             shardInformation.setLeaderVersion(leaderStateChanged.getLeaderPayloadVersion());
 764             if (shardInformation.setLeaderId(leaderStateChanged.getLeaderId())) {
 765                 primaryShardInfoCache.remove(shardInformation.getShardName());
 766             }
 767
 768             checkReady();
 769         } else {
 770             LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
 771         }
 772     }
 773
 774     private void onShardNotInitializedTimeout(final ShardNotInitializedTimeout message) {
 775         ShardInformation shardInfo = message.getShardInfo();
 776
 777         LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
 778                 shardInfo.getShardName());
 779
 780         shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
 781
 782         if (!shardInfo.isShardInitialized()) {
 783             LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
 784             message.getSender().tell(createNotInitializedException(shardInfo.getShardId()), getSelf());
 785         } else {
 786             LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
 787             message.getSender().tell(createNoShardLeaderException(shardInfo.getShardId()), getSelf());
 788         }
 789     }
 790
 791     private void onFollowerInitialSyncStatus(final FollowerInitialSyncUpStatus status) {
 792         LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
 793                 status.getName(), status.isInitialSyncDone());
 794
 795         ShardInformation shardInformation = findShardInformation(status.getName());
 796
 797         if (shardInformation != null) {
 798             shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
 799
 800             shardManagerMBean.setSyncStatus(isInSync());
 801         }
 802
 803     }
 804
 805     private void onRoleChangeNotification(final RoleChangeNotification roleChanged) {
 806         LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
 807                 roleChanged.getOldRole(), roleChanged.getNewRole());
 808
 809         ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
 810         if (shardInformation != null) {
 811             shardInformation.setRole(roleChanged.getNewRole());
 812             checkReady();
 813             shardManagerMBean.setSyncStatus(isInSync());
 814         }
 815     }
 816
 817
 818     private ShardInformation findShardInformation(final String memberId) {
 819         for (ShardInformation info : localShards.values()) {
 820             if (info.getShardId().toString().equals(memberId)) {
 821                 return info;
 822             }
 823         }
 824
 825         return null;
 826     }
 827
 828     private boolean isReadyWithLeaderId() {
 829         boolean isReady = true;
 830         for (ShardInformation info : localShards.values()) {
 831             if (!info.isShardReadyWithLeaderId()) {
 832                 isReady = false;
 833                 break;
 834             }
 835         }
 836         return isReady;
 837     }
 838
 839     private boolean isInSync() {
 840         for (ShardInformation info : localShards.values()) {
 841             if (!info.isInSync()) {
 842                 return false;
 843             }
 844         }
 845         return true;
 846     }
 847
 848     private void onActorInitialized(final Object message) {
 849         final ActorRef sender = getSender();
 850
 851         if (sender == null) {
 852             // why is a non-actor sending this message? Just ignore.
 853             return;
 854         }
 855
 856         String actorName = sender.path().name();
 857         //find shard name from actor name; actor name is stringified shardId
 858
 859         final ShardIdentifier shardId;
 860         try {
 861             shardId = ShardIdentifier.fromShardIdString(actorName);
 862         } catch (IllegalArgumentException e) {
 863             LOG.debug("{}: ignoring actor {}", actorName, e);
 864             return;
 865         }
 866
 867         markShardAsInitialized(shardId.getShardName());
 868     }
 869
 870     private void markShardAsInitialized(final String shardName) {
 871         LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
 872
 873         ShardInformation shardInformation = localShards.get(shardName);
 874         if (shardInformation != null) {
 875             shardInformation.setActorInitialized();
 876
 877             shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
 878         }
 879     }
 880
 881     @Override
 882     protected void handleRecover(final Object message) throws Exception {
 883         if (message instanceof RecoveryCompleted) {
 884             onRecoveryCompleted();
 885         } else if (message instanceof SnapshotOffer) {
 886             applyShardManagerSnapshot((ShardManagerSnapshot)((SnapshotOffer) message).snapshot());
 887         }
 888     }
 889
 890     @SuppressWarnings("checkstyle:IllegalCatch")
 891     private void onRecoveryCompleted() {
 892         LOG.info("Recovery complete : {}", persistenceId());
 893
 894         if (currentSnapshot == null && restoreFromSnapshot != null
 895                 && restoreFromSnapshot.getShardManagerSnapshot() != null) {
 896             ShardManagerSnapshot snapshot = restoreFromSnapshot.getShardManagerSnapshot();
 897
 898             LOG.debug("{}: Restoring from ShardManagerSnapshot: {}", persistenceId(), snapshot);
 899
 900             applyShardManagerSnapshot(snapshot);
 901         }
 902
 903         createLocalShards();
 904     }
 905
 906     private void sendResponse(final ShardInformation shardInformation, final boolean doWait,
 907             final boolean wantShardReady, final Supplier<Object> messageSupplier) {
 908         if (!shardInformation.isShardInitialized() || wantShardReady && !shardInformation.isShardReadyWithLeaderId()) {
 909             if (doWait) {
 910                 final ActorRef sender = getSender();
 911                 final ActorRef self = self();
 912
 913                 Runnable replyRunnable = () -> sender.tell(messageSupplier.get(), self);
 914
 915                 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
 916                     new OnShardInitialized(replyRunnable);
 917
 918                 shardInformation.addOnShardInitialized(onShardInitialized);
 919
 920                 FiniteDuration timeout = shardInformation.getDatastoreContext()
 921                         .getShardInitializationTimeout().duration();
 922                 if (shardInformation.isShardInitialized()) {
 923                     // If the shard is already initialized then we'll wait enough time for the shard to
 924                     // elect a leader, ie 2 times the election timeout.
 925                     timeout = FiniteDuration.create(shardInformation.getDatastoreContext().getShardRaftConfig()
 926                             .getElectionTimeOutInterval().toMillis() * 2, TimeUnit.MILLISECONDS);
 927                 }
 928
 929                 LOG.debug("{}: Scheduling {} ms timer to wait for shard {}", persistenceId(), timeout.toMillis(),
 930                         shardInformation.getShardName());
 931
 932                 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
 933                         timeout, getSelf(),
 934                         new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
 935                         getContext().dispatcher(), getSelf());
 936
 937                 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
 938
 939             } else if (!shardInformation.isShardInitialized()) {
 940                 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
 941                         shardInformation.getShardName());
 942                 getSender().tell(createNotInitializedException(shardInformation.getShardId()), getSelf());
 943             } else {
 944                 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
 945                         shardInformation.getShardName());
 946                 getSender().tell(createNoShardLeaderException(shardInformation.getShardId()), getSelf());
 947             }
 948
 949             return;
 950         }
 951
 952         getSender().tell(messageSupplier.get(), getSelf());
 953     }
 954
 955     private static NoShardLeaderException createNoShardLeaderException(final ShardIdentifier shardId) {
 956         return new NoShardLeaderException(null, shardId.toString());
 957     }
 958
 959     private static NotInitializedException createNotInitializedException(final ShardIdentifier shardId) {
 960         return new NotInitializedException(String.format(
 961                 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
 962     }
 963
 964     @VisibleForTesting
 965     static MemberName memberToName(final Member member) {
 966         return MemberName.forName(member.roles().iterator().next());
 967     }
 968
 969     private void memberRemoved(final ClusterEvent.MemberRemoved message) {
 970         MemberName memberName = memberToName(message.member());
 971
 972         LOG.info("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
 973                 message.member().address());
 974
 975         peerAddressResolver.removePeerAddress(memberName);
 976
 977         for (ShardInformation info : localShards.values()) {
 978             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 979         }
 980     }
 981
 982     private void memberExited(final ClusterEvent.MemberExited message) {
 983         MemberName memberName = memberToName(message.member());
 984
 985         LOG.info("{}: Received MemberExited: memberName: {}, address: {}", persistenceId(), memberName,
 986                 message.member().address());
 987
 988         peerAddressResolver.removePeerAddress(memberName);
 989
 990         for (ShardInformation info : localShards.values()) {
 991             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 992         }
 993     }
 994
 995     private void memberUp(final ClusterEvent.MemberUp message) {
 996         MemberName memberName = memberToName(message.member());
 997
 998         LOG.info("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
 999                 message.member().address());
1000
1001         memberUp(memberName, message.member().address());
1002     }
1003
1004     private void memberUp(final MemberName memberName, final Address address) {
1005         addPeerAddress(memberName, address);
1006         checkReady();
1007     }
1008
1009     private void memberWeaklyUp(final MemberWeaklyUp message) {
1010         MemberName memberName = memberToName(message.member());
1011
1012         LOG.info("{}: Received MemberWeaklyUp: memberName: {}, address: {}", persistenceId(), memberName,
1013                 message.member().address());
1014
1015         memberUp(memberName, message.member().address());
1016     }
1017
1018     private void addPeerAddress(final MemberName memberName, final Address address) {
1019         peerAddressResolver.addPeerAddress(memberName, address);
1020
1021         for (ShardInformation info : localShards.values()) {
1022             String shardName = info.getShardName();
1023             String peerId = getShardIdentifier(memberName, shardName).toString();
1024             info.updatePeerAddress(peerId, peerAddressResolver.getShardActorAddress(shardName, memberName), getSelf());
1025
1026             info.peerUp(memberName, peerId, getSelf());
1027         }
1028     }
1029
1030     private void memberReachable(final ClusterEvent.ReachableMember message) {
1031         MemberName memberName = memberToName(message.member());
1032         LOG.info("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
1033
1034         addPeerAddress(memberName, message.member().address());
1035
1036         markMemberAvailable(memberName);
1037     }
1038
1039     private void memberUnreachable(final ClusterEvent.UnreachableMember message) {
1040         MemberName memberName = memberToName(message.member());
1041         LOG.info("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
1042
1043         markMemberUnavailable(memberName);
1044     }
1045
1046     private void markMemberUnavailable(final MemberName memberName) {
1047         for (ShardInformation info : localShards.values()) {
1048             String leaderId = info.getLeaderId();
1049             if (leaderId != null && ShardIdentifier.fromShardIdString(leaderId).getMemberName().equals(memberName)) {
1050                 LOG.debug("Marking Leader {} as unavailable.", leaderId);
1051                 info.setLeaderAvailable(false);
1052
1053                 primaryShardInfoCache.remove(info.getShardName());
1054             }
1055
1056             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1057         }
1058     }
1059
1060     private void markMemberAvailable(final MemberName memberName) {
1061         for (ShardInformation info : localShards.values()) {
1062             String leaderId = info.getLeaderId();
1063             if (leaderId != null && ShardIdentifier.fromShardIdString(leaderId).getMemberName().equals(memberName)) {
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.getModules().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                 // Update peer address for every existing peer memeber to avoid missing sending
1137                 // PeerAddressResolved and PeerUp to this shard while UpdateSchemaContext comes after MemberUp.
1138                 String shardName = info.getShardName();
1139                 for (MemberName memberName : peerAddressResolver.getPeerMembers()) {
1140                     String peerId = getShardIdentifier(memberName, shardName).toString() ;
1141                     String peerAddress = peerAddressResolver.getShardActorAddress(shardName, memberName);
1142                     info.updatePeerAddress(peerId, peerAddress, getSelf());
1143                     info.peerUp(memberName, peerId, getSelf());
1144                     LOG.debug("{}: updated peer {} on member {} with address {} on shard {} whose actor address is {}",
1145                             persistenceId(), peerId, memberName, peerAddress, info.getShardId(), info.getActor());
1146                 }
1147             } else {
1148                 info.getActor().tell(message, getSelf());
1149             }
1150         }
1151     }
1152
1153     @VisibleForTesting
1154     protected ClusterWrapper getCluster() {
1155         return cluster;
1156     }
1157
1158     @VisibleForTesting
1159     protected ActorRef newShardActor(final ShardInformation info) {
1160         return getContext().actorOf(info.newProps().withDispatcher(shardDispatcherPath),
1161                 info.getShardId().toString());
1162     }
1163
1164     private void findPrimary(final FindPrimary message) {
1165         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
1166
1167         final String shardName = message.getShardName();
1168         final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
1169
1170         // First see if the there is a local replica for the shard
1171         final ShardInformation info = localShards.get(shardName);
1172         if (info != null && info.isActiveMember()) {
1173             sendResponse(info, message.isWaitUntilReady(), true, () -> {
1174                 String primaryPath = info.getSerializedLeaderActor();
1175                 Object found = canReturnLocalShardState && info.isLeader()
1176                         ? new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
1177                             new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
1178
1179                 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
1180                 return found;
1181             });
1182
1183             return;
1184         }
1185
1186         final Collection<String> visitedAddresses;
1187         if (message instanceof RemoteFindPrimary) {
1188             visitedAddresses = ((RemoteFindPrimary)message).getVisitedAddresses();
1189         } else {
1190             visitedAddresses = new ArrayList<>(1);
1191         }
1192
1193         visitedAddresses.add(peerAddressResolver.getShardManagerActorPathBuilder(cluster.getSelfAddress()).toString());
1194
1195         for (String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
1196             if (visitedAddresses.contains(address)) {
1197                 continue;
1198             }
1199
1200             LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}, visitedAddresses: {}",
1201                     persistenceId(), shardName, address, visitedAddresses);
1202
1203             getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
1204                     message.isWaitUntilReady(), visitedAddresses), getContext());
1205             return;
1206         }
1207
1208         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
1209
1210         getSender().tell(new PrimaryNotFoundException(
1211                 String.format("No primary shard found for %s.", shardName)), getSelf());
1212     }
1213
1214     private void findPrimary(final String shardName, final FindPrimaryResponseHandler handler) {
1215         Timeout findPrimaryTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1216                 .getShardInitializationTimeout().duration().$times(2));
1217
1218         Future<Object> futureObj = ask(getSelf(), new FindPrimary(shardName, true), findPrimaryTimeout);
1219         futureObj.onComplete(new OnComplete<Object>() {
1220             @Override
1221             public void onComplete(final Throwable failure, final Object response) {
1222                 if (failure != null) {
1223                     handler.onFailure(failure);
1224                 } else {
1225                     if (response instanceof RemotePrimaryShardFound) {
1226                         handler.onRemotePrimaryShardFound((RemotePrimaryShardFound) response);
1227                     } else if (response instanceof LocalPrimaryShardFound) {
1228                         handler.onLocalPrimaryFound((LocalPrimaryShardFound) response);
1229                     } else {
1230                         handler.onUnknownResponse(response);
1231                     }
1232                 }
1233             }
1234         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1235     }
1236
1237     /**
1238      * Construct the name of the shard actor given the name of the member on
1239      * which the shard resides and the name of the shard.
1240      *
1241      * @param memberName the member name
1242      * @param shardName the shard name
1243      * @return a b
1244      */
1245     private ShardIdentifier getShardIdentifier(final MemberName memberName, final String shardName) {
1246         return peerAddressResolver.getShardIdentifier(memberName, shardName);
1247     }
1248
1249     /**
1250      * Create shards that are local to the member on which the ShardManager runs.
1251      */
1252     private void createLocalShards() {
1253         MemberName memberName = this.cluster.getCurrentMemberName();
1254         Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
1255
1256         Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots = new HashMap<>();
1257         if (restoreFromSnapshot != null) {
1258             for (DatastoreSnapshot.ShardSnapshot snapshot: restoreFromSnapshot.getShardSnapshots()) {
1259                 shardSnapshots.put(snapshot.getName(), snapshot);
1260             }
1261         }
1262
1263         // null out to GC
1264         restoreFromSnapshot = null;
1265
1266         for (String shardName : memberShardNames) {
1267             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1268
1269             LOG.debug("{}: Creating local shard: {}", persistenceId(), shardId);
1270
1271             Map<String, String> peerAddresses = getPeerAddresses(shardName);
1272             localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses,
1273                     newShardDatastoreContext(shardName), Shard.builder().restoreFromSnapshot(
1274                         shardSnapshots.get(shardName)), peerAddressResolver));
1275         }
1276     }
1277
1278     /**
1279      * Given the name of the shard find the addresses of all it's peers.
1280      *
1281      * @param shardName the shard name
1282      */
1283     private Map<String, String> getPeerAddresses(final String shardName) {
1284         final Collection<MemberName> members = configuration.getMembersFromShardName(shardName);
1285         return getPeerAddresses(shardName, members);
1286     }
1287
1288     private Map<String, String> getPeerAddresses(final String shardName, final Collection<MemberName> members) {
1289         Map<String, String> peerAddresses = new HashMap<>();
1290         MemberName currentMemberName = this.cluster.getCurrentMemberName();
1291
1292         for (MemberName memberName : members) {
1293             if (!currentMemberName.equals(memberName)) {
1294                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1295                 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
1296                 peerAddresses.put(shardId.toString(), address);
1297             }
1298         }
1299         return peerAddresses;
1300     }
1301
1302     @Override
1303     public SupervisorStrategy supervisorStrategy() {
1304
1305         return new OneForOneStrategy(10, Duration.create("1 minute"),
1306                 (Function<Throwable, Directive>) t -> {
1307                     LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
1308                     return SupervisorStrategy.resume();
1309                 });
1310     }
1311
1312     @Override
1313     public String persistenceId() {
1314         return persistenceId;
1315     }
1316
1317     @VisibleForTesting
1318     ShardManagerInfoMBean getMBean() {
1319         return shardManagerMBean;
1320     }
1321
1322     private boolean isShardReplicaOperationInProgress(final String shardName, final ActorRef sender) {
1323         if (shardReplicaOperationsInProgress.contains(shardName)) {
1324             String msg = String.format("A shard replica operation for %s is already in progress", shardName);
1325             LOG.debug("{}: {}", persistenceId(), msg);
1326             sender.tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1327             return true;
1328         }
1329
1330         return false;
1331     }
1332
1333     private void onAddPrefixShardReplica(final AddPrefixShardReplica message) {
1334         LOG.debug("{}: onAddPrefixShardReplica: {}", persistenceId(), message);
1335
1336         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1337                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1338         final String shardName = shardId.getShardName();
1339
1340         // Create the localShard
1341         if (schemaContext == null) {
1342             String msg = String.format(
1343                     "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1344             LOG.debug("{}: {}", persistenceId(), msg);
1345             getSender().tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1346             return;
1347         }
1348
1349         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1350                 getSelf()) {
1351             @Override
1352             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1353                 final RunnableMessage runnable = (RunnableMessage) () -> addPrefixShard(getShardName(),
1354                         message.getShardPrefix(), response, getSender());
1355                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1356                     getSelf().tell(runnable, getTargetActor());
1357                 }
1358             }
1359
1360             @Override
1361             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1362                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1363             }
1364         });
1365     }
1366
1367     private void onAddShardReplica(final AddShardReplica shardReplicaMsg) {
1368         final String shardName = shardReplicaMsg.getShardName();
1369
1370         LOG.debug("{}: onAddShardReplica: {}", persistenceId(), shardReplicaMsg);
1371
1372         // verify the shard with the specified name is present in the cluster configuration
1373         if (!this.configuration.isShardConfigured(shardName)) {
1374             String msg = String.format("No module configuration exists for shard %s", shardName);
1375             LOG.debug("{}: {}", persistenceId(), msg);
1376             getSender().tell(new Status.Failure(new IllegalArgumentException(msg)), getSelf());
1377             return;
1378         }
1379
1380         // Create the localShard
1381         if (schemaContext == null) {
1382             String msg = String.format(
1383                   "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1384             LOG.debug("{}: {}", persistenceId(), msg);
1385             getSender().tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1386             return;
1387         }
1388
1389         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1390                 getSelf()) {
1391             @Override
1392             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1393                 final RunnableMessage runnable = (RunnableMessage) () ->
1394                     addShard(getShardName(), response, getSender());
1395                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1396                     getSelf().tell(runnable, getTargetActor());
1397                 }
1398             }
1399
1400             @Override
1401             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1402                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1403             }
1404         });
1405     }
1406
1407     private void sendLocalReplicaAlreadyExistsReply(final String shardName, final ActorRef sender) {
1408         String msg = String.format("Local shard %s already exists", shardName);
1409         LOG.debug("{}: {}", persistenceId(), msg);
1410         sender.tell(new Status.Failure(new AlreadyExistsException(msg)), getSelf());
1411     }
1412
1413     private void addPrefixShard(final String shardName, final YangInstanceIdentifier shardPrefix,
1414                                 final RemotePrimaryShardFound response, final ActorRef sender) {
1415         if (isShardReplicaOperationInProgress(shardName, sender)) {
1416             return;
1417         }
1418
1419         shardReplicaOperationsInProgress.add(shardName);
1420
1421         final ShardInformation shardInfo;
1422         final boolean removeShardOnFailure;
1423         ShardInformation existingShardInfo = localShards.get(shardName);
1424         if (existingShardInfo == null) {
1425             removeShardOnFailure = true;
1426             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1427
1428             final Builder builder = newShardDatastoreContextBuilder(shardName);
1429             builder.storeRoot(shardPrefix).customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName());
1430
1431             DatastoreContext datastoreContext = builder.build();
1432
1433             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1434                     Shard.builder(), peerAddressResolver);
1435             shardInfo.setActiveMember(false);
1436             shardInfo.setSchemaContext(schemaContext);
1437             localShards.put(shardName, shardInfo);
1438             shardInfo.setActor(newShardActor(shardInfo));
1439         } else {
1440             removeShardOnFailure = false;
1441             shardInfo = existingShardInfo;
1442         }
1443
1444         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1445     }
1446
1447     private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) {
1448         if (isShardReplicaOperationInProgress(shardName, sender)) {
1449             return;
1450         }
1451
1452         shardReplicaOperationsInProgress.add(shardName);
1453
1454         final ShardInformation shardInfo;
1455         final boolean removeShardOnFailure;
1456         ShardInformation existingShardInfo = localShards.get(shardName);
1457         if (existingShardInfo == null) {
1458             removeShardOnFailure = true;
1459             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1460
1461             DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName)
1462                     .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
1463
1464             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1465                     Shard.builder(), peerAddressResolver);
1466             shardInfo.setActiveMember(false);
1467             shardInfo.setSchemaContext(schemaContext);
1468             localShards.put(shardName, shardInfo);
1469             shardInfo.setActor(newShardActor(shardInfo));
1470         } else {
1471             removeShardOnFailure = false;
1472             shardInfo = existingShardInfo;
1473         }
1474
1475         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1476     }
1477
1478     private void execAddShard(final String shardName,
1479                               final ShardInformation shardInfo,
1480                               final RemotePrimaryShardFound response,
1481                               final boolean removeShardOnFailure,
1482                               final ActorRef sender) {
1483
1484         final String localShardAddress =
1485                 peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName());
1486
1487         //inform ShardLeader to add this shard as a replica by sending an AddServer message
1488         LOG.debug("{}: Sending AddServer message to peer {} for shard {}", persistenceId(),
1489                 response.getPrimaryPath(), shardInfo.getShardId());
1490
1491         final Timeout addServerTimeout = new Timeout(shardInfo.getDatastoreContext()
1492                 .getShardLeaderElectionTimeout().duration());
1493         final Future<Object> futureObj = ask(getContext().actorSelection(response.getPrimaryPath()),
1494                 new AddServer(shardInfo.getShardId().toString(), localShardAddress, true), addServerTimeout);
1495
1496         futureObj.onComplete(new OnComplete<Object>() {
1497             @Override
1498             public void onComplete(final Throwable failure, final Object addServerResponse) {
1499                 if (failure != null) {
1500                     LOG.debug("{}: AddServer request to {} for {} failed", persistenceId(),
1501                             response.getPrimaryPath(), shardName, failure);
1502
1503                     final String msg = String.format("AddServer request to leader %s for shard %s failed",
1504                             response.getPrimaryPath(), shardName);
1505                     self().tell(new ForwardedAddServerFailure(shardName, msg, failure, removeShardOnFailure), sender);
1506                 } else {
1507                     self().tell(new ForwardedAddServerReply(shardInfo, (AddServerReply)addServerResponse,
1508                             response.getPrimaryPath(), removeShardOnFailure), sender);
1509                 }
1510             }
1511         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1512     }
1513
1514     private void onAddServerFailure(final String shardName, final String message, final Throwable failure,
1515             final ActorRef sender, final boolean removeShardOnFailure) {
1516         shardReplicaOperationsInProgress.remove(shardName);
1517
1518         if (removeShardOnFailure) {
1519             ShardInformation shardInfo = localShards.remove(shardName);
1520             if (shardInfo.getActor() != null) {
1521                 shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf());
1522             }
1523         }
1524
1525         sender.tell(new Status.Failure(message == null ? failure :
1526             new RuntimeException(message, failure)), getSelf());
1527     }
1528
1529     private void onAddServerReply(final ShardInformation shardInfo, final AddServerReply replyMsg,
1530             final ActorRef sender, final String leaderPath, final boolean removeShardOnFailure) {
1531         String shardName = shardInfo.getShardName();
1532         shardReplicaOperationsInProgress.remove(shardName);
1533
1534         LOG.debug("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath);
1535
1536         if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1537             LOG.debug("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName);
1538
1539             // Make the local shard voting capable
1540             shardInfo.setDatastoreContext(newShardDatastoreContext(shardName), getSelf());
1541             shardInfo.setActiveMember(true);
1542             persistShardList();
1543
1544             sender.tell(new Status.Success(null), getSelf());
1545         } else if (replyMsg.getStatus() == ServerChangeStatus.ALREADY_EXISTS) {
1546             sendLocalReplicaAlreadyExistsReply(shardName, sender);
1547         } else {
1548             LOG.warn("{}: Leader failed to add shard replica {} with status {}",
1549                     persistenceId(), shardName, replyMsg.getStatus());
1550
1551             Exception failure = getServerChangeException(AddServer.class, replyMsg.getStatus(), leaderPath,
1552                     shardInfo.getShardId());
1553
1554             onAddServerFailure(shardName, null, failure, sender, removeShardOnFailure);
1555         }
1556     }
1557
1558     private static Exception getServerChangeException(final Class<?> serverChange,
1559             final ServerChangeStatus serverChangeStatus, final String leaderPath, final ShardIdentifier shardId) {
1560         switch (serverChangeStatus) {
1561             case TIMEOUT:
1562                 return new TimeoutException(String.format(
1563                         "The shard leader %s timed out trying to replicate the initial data to the new shard %s."
1564                         + "Possible causes - there was a problem replicating the data or shard leadership changed "
1565                         + "while replicating the shard data", leaderPath, shardId.getShardName()));
1566             case NO_LEADER:
1567                 return createNoShardLeaderException(shardId);
1568             case NOT_SUPPORTED:
1569                 return new UnsupportedOperationException(String.format("%s request is not supported for shard %s",
1570                         serverChange.getSimpleName(), shardId.getShardName()));
1571             default :
1572                 return new RuntimeException(String.format("%s request to leader %s for shard %s failed with status %s",
1573                         serverChange.getSimpleName(), leaderPath, shardId.getShardName(), serverChangeStatus));
1574         }
1575     }
1576
1577     private void onRemoveShardReplica(final RemoveShardReplica shardReplicaMsg) {
1578         LOG.debug("{}: onRemoveShardReplica: {}", persistenceId(), shardReplicaMsg);
1579
1580         findPrimary(shardReplicaMsg.getShardName(), new AutoFindPrimaryFailureResponseHandler(getSender(),
1581                 shardReplicaMsg.getShardName(), persistenceId(), getSelf()) {
1582             @Override
1583             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1584                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1585             }
1586
1587             @Override
1588             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1589                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1590             }
1591
1592             private void doRemoveShardReplicaAsync(final String primaryPath) {
1593                 getSelf().tell((RunnableMessage) () -> removeShardReplica(shardReplicaMsg, getShardName(),
1594                         primaryPath, getSender()), getTargetActor());
1595             }
1596         });
1597     }
1598
1599     private void onRemovePrefixShardReplica(final RemovePrefixShardReplica message) {
1600         LOG.debug("{}: onRemovePrefixShardReplica: {}", persistenceId(), message);
1601
1602         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1603                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1604         final String shardName = shardId.getShardName();
1605
1606         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(),
1607                 shardName, persistenceId(), getSelf()) {
1608             @Override
1609             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1610                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1611             }
1612
1613             @Override
1614             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1615                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1616             }
1617
1618             private void doRemoveShardReplicaAsync(final String primaryPath) {
1619                 getSelf().tell((RunnableMessage) () -> removePrefixShardReplica(message, getShardName(),
1620                         primaryPath, getSender()), getTargetActor());
1621             }
1622         });
1623     }
1624
1625     private void persistShardList() {
1626         List<String> shardList = new ArrayList<>(localShards.keySet());
1627         for (ShardInformation shardInfo : localShards.values()) {
1628             if (!shardInfo.isActiveMember()) {
1629                 shardList.remove(shardInfo.getShardName());
1630             }
1631         }
1632         LOG.debug("{}: persisting the shard list {}", persistenceId(), shardList);
1633         saveSnapshot(updateShardManagerSnapshot(shardList, configuration.getAllPrefixShardConfigurations()));
1634     }
1635
1636     private ShardManagerSnapshot updateShardManagerSnapshot(
1637             final List<String> shardList,
1638             final Map<DOMDataTreeIdentifier, PrefixShardConfiguration> allPrefixShardConfigurations) {
1639         currentSnapshot = new ShardManagerSnapshot(shardList, allPrefixShardConfigurations);
1640         return currentSnapshot;
1641     }
1642
1643     private void applyShardManagerSnapshot(final ShardManagerSnapshot snapshot) {
1644         currentSnapshot = snapshot;
1645
1646         LOG.debug("{}: onSnapshotOffer: {}", persistenceId(), currentSnapshot);
1647
1648         final MemberName currentMember = cluster.getCurrentMemberName();
1649         Set<String> configuredShardList =
1650             new HashSet<>(configuration.getMemberShardNames(currentMember));
1651         for (String shard : currentSnapshot.getShardList()) {
1652             if (!configuredShardList.contains(shard)) {
1653                 // add the current member as a replica for the shard
1654                 LOG.debug("{}: adding shard {}", persistenceId(), shard);
1655                 configuration.addMemberReplicaForShard(shard, currentMember);
1656             } else {
1657                 configuredShardList.remove(shard);
1658             }
1659         }
1660         for (String shard : configuredShardList) {
1661             // remove the member as a replica for the shard
1662             LOG.debug("{}: removing shard {}", persistenceId(), shard);
1663             configuration.removeMemberReplicaForShard(shard, currentMember);
1664         }
1665     }
1666
1667     private void onSaveSnapshotSuccess(final SaveSnapshotSuccess successMessage) {
1668         LOG.debug("{} saved ShardManager snapshot successfully. Deleting the prev snapshot if available",
1669             persistenceId());
1670         deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), successMessage.metadata().timestamp() - 1,
1671             0, 0));
1672     }
1673
1674     private void onChangeShardServersVotingStatus(final ChangeShardMembersVotingStatus changeMembersVotingStatus) {
1675         LOG.debug("{}: onChangeShardServersVotingStatus: {}", persistenceId(), changeMembersVotingStatus);
1676
1677         String shardName = changeMembersVotingStatus.getShardName();
1678         Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1679         for (Entry<String, Boolean> e: changeMembersVotingStatus.getMeberVotingStatusMap().entrySet()) {
1680             serverVotingStatusMap.put(getShardIdentifier(MemberName.forName(e.getKey()), shardName).toString(),
1681                     e.getValue());
1682         }
1683
1684         ChangeServersVotingStatus changeServersVotingStatus = new ChangeServersVotingStatus(serverVotingStatusMap);
1685
1686         findLocalShard(shardName, getSender(),
1687             localShardFound -> changeShardMembersVotingStatus(changeServersVotingStatus, shardName,
1688             localShardFound.getPath(), getSender()));
1689     }
1690
1691     private void onFlipShardMembersVotingStatus(final FlipShardMembersVotingStatus flipMembersVotingStatus) {
1692         LOG.debug("{}: onFlipShardMembersVotingStatus: {}", persistenceId(), flipMembersVotingStatus);
1693
1694         ActorRef sender = getSender();
1695         final String shardName = flipMembersVotingStatus.getShardName();
1696         findLocalShard(shardName, sender, localShardFound -> {
1697             Future<Object> future = ask(localShardFound.getPath(), GetOnDemandRaftState.INSTANCE,
1698                     Timeout.apply(30, TimeUnit.SECONDS));
1699
1700             future.onComplete(new OnComplete<Object>() {
1701                 @Override
1702                 public void onComplete(final Throwable failure, final Object response) {
1703                     if (failure != null) {
1704                         sender.tell(new Status.Failure(new RuntimeException(
1705                                 String.format("Failed to access local shard %s", shardName), failure)), self());
1706                         return;
1707                     }
1708
1709                     OnDemandRaftState raftState = (OnDemandRaftState) response;
1710                     Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1711                     for (Entry<String, Boolean> e: raftState.getPeerVotingStates().entrySet()) {
1712                         serverVotingStatusMap.put(e.getKey(), !e.getValue());
1713                     }
1714
1715                     serverVotingStatusMap.put(getShardIdentifier(cluster.getCurrentMemberName(), shardName)
1716                             .toString(), !raftState.isVoting());
1717
1718                     changeShardMembersVotingStatus(new ChangeServersVotingStatus(serverVotingStatusMap),
1719                             shardName, localShardFound.getPath(), sender);
1720                 }
1721             }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1722         });
1723
1724     }
1725
1726     private void findLocalShard(final FindLocalShard message) {
1727         LOG.debug("{}: findLocalShard : {}", persistenceId(), message.getShardName());
1728
1729         final ShardInformation shardInformation = localShards.get(message.getShardName());
1730
1731         if (shardInformation == null) {
1732             LOG.debug("{}: Local shard {} not found - shards present: {}",
1733                     persistenceId(), message.getShardName(), localShards.keySet());
1734
1735             getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
1736             return;
1737         }
1738
1739         sendResponse(shardInformation, message.isWaitUntilInitialized(), false,
1740             () -> new LocalShardFound(shardInformation.getActor()));
1741     }
1742
1743     private void findLocalShard(final String shardName, final ActorRef sender,
1744             final Consumer<LocalShardFound> onLocalShardFound) {
1745         Timeout findLocalTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1746                 .getShardInitializationTimeout().duration().$times(2));
1747
1748         Future<Object> futureObj = ask(getSelf(), new FindLocalShard(shardName, true), findLocalTimeout);
1749         futureObj.onComplete(new OnComplete<Object>() {
1750             @Override
1751             public void onComplete(final Throwable failure, final Object response) {
1752                 if (failure != null) {
1753                     LOG.debug("{}: Received failure from FindLocalShard for shard {}", persistenceId, shardName,
1754                             failure);
1755                     sender.tell(new Status.Failure(new RuntimeException(
1756                             String.format("Failed to find local shard %s", shardName), failure)), self());
1757                 } else {
1758                     if (response instanceof LocalShardFound) {
1759                         getSelf().tell((RunnableMessage) () -> onLocalShardFound.accept((LocalShardFound) response),
1760                                 sender);
1761                     } else if (response instanceof LocalShardNotFound) {
1762                         String msg = String.format("Local shard %s does not exist", shardName);
1763                         LOG.debug("{}: {}", persistenceId, msg);
1764                         sender.tell(new Status.Failure(new IllegalArgumentException(msg)), self());
1765                     } else {
1766                         String msg = String.format("Failed to find local shard %s: received response: %s",
1767                                 shardName, response);
1768                         LOG.debug("{}: {}", persistenceId, msg);
1769                         sender.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response :
1770                                 new RuntimeException(msg)), self());
1771                     }
1772                 }
1773             }
1774         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1775     }
1776
1777     private void changeShardMembersVotingStatus(final ChangeServersVotingStatus changeServersVotingStatus,
1778             final String shardName, final ActorRef shardActorRef, final ActorRef sender) {
1779         if (isShardReplicaOperationInProgress(shardName, sender)) {
1780             return;
1781         }
1782
1783         shardReplicaOperationsInProgress.add(shardName);
1784
1785         DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
1786         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1787
1788         LOG.debug("{}: Sending ChangeServersVotingStatus message {} to local shard {}", persistenceId(),
1789                 changeServersVotingStatus, shardActorRef.path());
1790
1791         Timeout timeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().duration().$times(2));
1792         Future<Object> futureObj = ask(shardActorRef, changeServersVotingStatus, timeout);
1793
1794         futureObj.onComplete(new OnComplete<Object>() {
1795             @Override
1796             public void onComplete(final Throwable failure, final Object response) {
1797                 shardReplicaOperationsInProgress.remove(shardName);
1798                 if (failure != null) {
1799                     String msg = String.format("ChangeServersVotingStatus request to local shard %s failed",
1800                             shardActorRef.path());
1801                     LOG.debug("{}: {}", persistenceId(), msg, failure);
1802                     sender.tell(new Status.Failure(new RuntimeException(msg, failure)), self());
1803                 } else {
1804                     LOG.debug("{}: Received {} from local shard {}", persistenceId(), response, shardActorRef.path());
1805
1806                     ServerChangeReply replyMsg = (ServerChangeReply) response;
1807                     if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1808                         LOG.debug("{}: ChangeServersVotingStatus succeeded for shard {}", persistenceId(), shardName);
1809                         sender.tell(new Status.Success(null), getSelf());
1810                     } else if (replyMsg.getStatus() == ServerChangeStatus.INVALID_REQUEST) {
1811                         sender.tell(new Status.Failure(new IllegalArgumentException(String.format(
1812                                 "The requested voting state change for shard %s is invalid. At least one member "
1813                                 + "must be voting", shardId.getShardName()))), getSelf());
1814                     } else {
1815                         LOG.warn("{}: ChangeServersVotingStatus failed for shard {} with status {}",
1816                                 persistenceId(), shardName, replyMsg.getStatus());
1817
1818                         Exception error = getServerChangeException(ChangeServersVotingStatus.class,
1819                                 replyMsg.getStatus(), shardActorRef.path().toString(), shardId);
1820                         sender.tell(new Status.Failure(error), getSelf());
1821                     }
1822                 }
1823             }
1824         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1825     }
1826
1827     private static final class ForwardedAddServerReply {
1828         ShardInformation shardInfo;
1829         AddServerReply addServerReply;
1830         String leaderPath;
1831         boolean removeShardOnFailure;
1832
1833         ForwardedAddServerReply(final ShardInformation shardInfo, final AddServerReply addServerReply,
1834             final String leaderPath, final boolean removeShardOnFailure) {
1835             this.shardInfo = shardInfo;
1836             this.addServerReply = addServerReply;
1837             this.leaderPath = leaderPath;
1838             this.removeShardOnFailure = removeShardOnFailure;
1839         }
1840     }
1841
1842     private static final class ForwardedAddServerFailure {
1843         String shardName;
1844         String failureMessage;
1845         Throwable failure;
1846         boolean removeShardOnFailure;
1847
1848         ForwardedAddServerFailure(final String shardName, final String failureMessage, final Throwable failure,
1849                 final boolean removeShardOnFailure) {
1850             this.shardName = shardName;
1851             this.failureMessage = failureMessage;
1852             this.failure = failure;
1853             this.removeShardOnFailure = removeShardOnFailure;
1854         }
1855     }
1856
1857     static class OnShardInitialized {
1858         private final Runnable replyRunnable;
1859         private Cancellable timeoutSchedule;
1860
1861         OnShardInitialized(final Runnable replyRunnable) {
1862             this.replyRunnable = replyRunnable;
1863         }
1864
1865         Runnable getReplyRunnable() {
1866             return replyRunnable;
1867         }
1868
1869         Cancellable getTimeoutSchedule() {
1870             return timeoutSchedule;
1871         }
1872
1873         void setTimeoutSchedule(final Cancellable timeoutSchedule) {
1874             this.timeoutSchedule = timeoutSchedule;
1875         }
1876     }
1877
1878     static class OnShardReady extends OnShardInitialized {
1879         OnShardReady(final Runnable replyRunnable) {
1880             super(replyRunnable);
1881         }
1882     }
1883
1884     private interface RunnableMessage extends Runnable {
1885     }
1886
1887     /**
1888      * The FindPrimaryResponseHandler provides specific callback methods which are invoked when a response to the
1889      * a remote or local find primary message is processed.
1890      */
1891     private interface FindPrimaryResponseHandler {
1892         /**
1893          * Invoked when a Failure message is received as a response.
1894          *
1895          * @param failure the failure exception
1896          */
1897         void onFailure(Throwable failure);
1898
1899         /**
1900          * Invoked when a RemotePrimaryShardFound response is received.
1901          *
1902          * @param response the response
1903          */
1904         void onRemotePrimaryShardFound(RemotePrimaryShardFound response);
1905
1906         /**
1907          * Invoked when a LocalPrimaryShardFound response is received.
1908          *
1909          * @param response the response
1910          */
1911         void onLocalPrimaryFound(LocalPrimaryShardFound response);
1912
1913         /**
1914          * Invoked when an unknown response is received. This is another type of failure.
1915          *
1916          * @param response the response
1917          */
1918         void onUnknownResponse(Object response);
1919     }
1920
1921     /**
1922      * The AutoFindPrimaryFailureResponseHandler automatically processes Failure responses when finding a primary
1923      * replica and sends a wrapped Failure response to some targetActor.
1924      */
1925     private abstract static class AutoFindPrimaryFailureResponseHandler implements FindPrimaryResponseHandler {
1926         private final ActorRef targetActor;
1927         private final String shardName;
1928         private final String persistenceId;
1929         private final ActorRef shardManagerActor;
1930
1931         /**
1932          * Constructs an instance.
1933          *
1934          * @param targetActor The actor to whom the Failure response should be sent when a FindPrimary failure occurs
1935          * @param shardName The name of the shard for which the primary replica had to be found
1936          * @param persistenceId The persistenceId for the ShardManager
1937          * @param shardManagerActor The ShardManager actor which triggered the call to FindPrimary
1938          */
1939         protected AutoFindPrimaryFailureResponseHandler(final ActorRef targetActor, final String shardName,
1940                 final String persistenceId, final ActorRef shardManagerActor) {
1941             this.targetActor = Preconditions.checkNotNull(targetActor);
1942             this.shardName = Preconditions.checkNotNull(shardName);
1943             this.persistenceId = Preconditions.checkNotNull(persistenceId);
1944             this.shardManagerActor = Preconditions.checkNotNull(shardManagerActor);
1945         }
1946
1947         public ActorRef getTargetActor() {
1948             return targetActor;
1949         }
1950
1951         public String getShardName() {
1952             return shardName;
1953         }
1954
1955         @Override
1956         public void onFailure(final Throwable failure) {
1957             LOG.debug("{}: Received failure from FindPrimary for shard {}", persistenceId, shardName, failure);
1958             targetActor.tell(new Status.Failure(new RuntimeException(
1959                     String.format("Failed to find leader for shard %s", shardName), failure)), shardManagerActor);
1960         }
1961
1962         @Override
1963         public void onUnknownResponse(final Object response) {
1964             String msg = String.format("Failed to find leader for shard %s: received response: %s",
1965                     shardName, response);
1966             LOG.debug("{}: {}", persistenceId, msg);
1967             targetActor.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response :
1968                     new RuntimeException(msg)), shardManagerActor);
1969         }
1970     }
1971
1972     /**
1973      * The WrappedShardResponse class wraps a response from a Shard.
1974      */
1975     private static final class WrappedShardResponse {
1976         private final ShardIdentifier shardId;
1977         private final Object response;
1978         private final String leaderPath;
1979
1980         WrappedShardResponse(final ShardIdentifier shardId, final Object response, final String leaderPath) {
1981             this.shardId = shardId;
1982             this.response = response;
1983             this.leaderPath = leaderPath;
1984         }
1985
1986         ShardIdentifier getShardId() {
1987             return shardId;
1988         }
1989
1990         Object getResponse() {
1991             return response;
1992         }
1993
1994         String getLeaderPath() {
1995             return leaderPath;
1996         }
1997     }
1998
1999     private static final class ShardNotInitializedTimeout {
2000         private final ActorRef sender;
2001         private final ShardInformation shardInfo;
2002         private final OnShardInitialized onShardInitialized;
2003
2004         ShardNotInitializedTimeout(final ShardInformation shardInfo, final OnShardInitialized onShardInitialized,
2005             final ActorRef sender) {
2006             this.sender = sender;
2007             this.shardInfo = shardInfo;
2008             this.onShardInitialized = onShardInitialized;
2009         }
2010
2011         ActorRef getSender() {
2012             return sender;
2013         }
2014
2015         ShardInformation getShardInfo() {
2016             return shardInfo;
2017         }
2018
2019         OnShardInitialized getOnShardInitialized() {
2020             return onShardInitialized;
2021         }
2022     }
2023 }
2024
2025
2026