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         final String memberStr = memberName.getName();
1048         for (ShardInformation info : localShards.values()) {
1049             String leaderId = info.getLeaderId();
1050             // XXX: why are we using String#contains() here?
1051             if (leaderId != null && leaderId.contains(memberStr)) {
1052                 LOG.debug("Marking Leader {} as unavailable.", leaderId);
1053                 info.setLeaderAvailable(false);
1054
1055                 primaryShardInfoCache.remove(info.getShardName());
1056             }
1057
1058             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1059         }
1060     }
1061
1062     private void markMemberAvailable(final MemberName memberName) {
1063         final String memberStr = memberName.getName();
1064         for (ShardInformation info : localShards.values()) {
1065             String leaderId = info.getLeaderId();
1066             // XXX: why are we using String#contains() here?
1067             if (leaderId != null && leaderId.contains(memberStr)) {
1068                 LOG.debug("Marking Leader {} as available.", leaderId);
1069                 info.setLeaderAvailable(true);
1070             }
1071
1072             info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
1073         }
1074     }
1075
1076     private void onDatastoreContextFactory(final DatastoreContextFactory factory) {
1077         datastoreContextFactory = factory;
1078         for (ShardInformation info : localShards.values()) {
1079             info.setDatastoreContext(newShardDatastoreContext(info.getShardName()), getSelf());
1080         }
1081     }
1082
1083     private void onGetLocalShardIds() {
1084         final List<String> response = new ArrayList<>(localShards.size());
1085
1086         for (ShardInformation info : localShards.values()) {
1087             response.add(info.getShardId().toString());
1088         }
1089
1090         getSender().tell(new Status.Success(response), getSelf());
1091     }
1092
1093     private void onSwitchShardBehavior(final SwitchShardBehavior message) {
1094         final ShardIdentifier identifier = message.getShardId();
1095
1096         if (identifier != null) {
1097             final ShardInformation info = localShards.get(identifier.getShardName());
1098             if (info == null) {
1099                 getSender().tell(new Status.Failure(
1100                     new IllegalArgumentException("Shard " + identifier + " is not local")), getSelf());
1101                 return;
1102             }
1103
1104             switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
1105         } else {
1106             for (ShardInformation info : localShards.values()) {
1107                 switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
1108             }
1109         }
1110
1111         getSender().tell(new Status.Success(null), getSelf());
1112     }
1113
1114     private void switchShardBehavior(final ShardInformation info, final SwitchBehavior switchBehavior) {
1115         final ActorRef actor = info.getActor();
1116         if (actor != null) {
1117             actor.tell(switchBehavior, getSelf());
1118         } else {
1119             LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
1120                 info.getShardName(), switchBehavior.getNewState());
1121         }
1122     }
1123
1124     /**
1125      * Notifies all the local shards of a change in the schema context.
1126      *
1127      * @param message the message to send
1128      */
1129     private void updateSchemaContext(final Object message) {
1130         schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
1131
1132         LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getAllModuleIdentifiers().size());
1133
1134         for (ShardInformation info : localShards.values()) {
1135             info.setSchemaContext(schemaContext);
1136
1137             if (info.getActor() == null) {
1138                 LOG.debug("Creating Shard {}", info.getShardId());
1139                 info.setActor(newShardActor(info));
1140                 // Update peer address for every existing peer memeber to avoid missing sending
1141                 // PeerAddressResolved and PeerUp to this shard while UpdateSchemaContext comes after MemberUp.
1142                 String shardName = info.getShardName();
1143                 for (MemberName memberName : peerAddressResolver.getPeerMembers()) {
1144                     String peerId = getShardIdentifier(memberName, shardName).toString() ;
1145                     String peerAddress = peerAddressResolver.getShardActorAddress(shardName, memberName);
1146                     info.updatePeerAddress(peerId, peerAddress, getSelf());
1147                     info.peerUp(memberName, peerId, getSelf());
1148                     LOG.debug("{}: updated peer {} on member {} with address {} on shard {} whose actor address is {}",
1149                             persistenceId(), peerId, memberName, peerAddress, info.getShardId(), info.getActor());
1150                 }
1151             } else {
1152                 info.getActor().tell(message, getSelf());
1153             }
1154         }
1155     }
1156
1157     @VisibleForTesting
1158     protected ClusterWrapper getCluster() {
1159         return cluster;
1160     }
1161
1162     @VisibleForTesting
1163     protected ActorRef newShardActor(final ShardInformation info) {
1164         return getContext().actorOf(info.newProps().withDispatcher(shardDispatcherPath),
1165                 info.getShardId().toString());
1166     }
1167
1168     private void findPrimary(final FindPrimary message) {
1169         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
1170
1171         final String shardName = message.getShardName();
1172         final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
1173
1174         // First see if the there is a local replica for the shard
1175         final ShardInformation info = localShards.get(shardName);
1176         if (info != null && info.isActiveMember()) {
1177             sendResponse(info, message.isWaitUntilReady(), true, () -> {
1178                 String primaryPath = info.getSerializedLeaderActor();
1179                 Object found = canReturnLocalShardState && info.isLeader()
1180                         ? new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
1181                             new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
1182
1183                 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
1184                 return found;
1185             });
1186
1187             return;
1188         }
1189
1190         final Collection<String> visitedAddresses;
1191         if (message instanceof RemoteFindPrimary) {
1192             visitedAddresses = ((RemoteFindPrimary)message).getVisitedAddresses();
1193         } else {
1194             visitedAddresses = new ArrayList<>(1);
1195         }
1196
1197         visitedAddresses.add(peerAddressResolver.getShardManagerActorPathBuilder(cluster.getSelfAddress()).toString());
1198
1199         for (String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
1200             if (visitedAddresses.contains(address)) {
1201                 continue;
1202             }
1203
1204             LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}, visitedAddresses: {}",
1205                     persistenceId(), shardName, address, visitedAddresses);
1206
1207             getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
1208                     message.isWaitUntilReady(), visitedAddresses), getContext());
1209             return;
1210         }
1211
1212         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
1213
1214         getSender().tell(new PrimaryNotFoundException(
1215                 String.format("No primary shard found for %s.", shardName)), getSelf());
1216     }
1217
1218     private void findPrimary(final String shardName, final FindPrimaryResponseHandler handler) {
1219         Timeout findPrimaryTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1220                 .getShardInitializationTimeout().duration().$times(2));
1221
1222         Future<Object> futureObj = ask(getSelf(), new FindPrimary(shardName, true), findPrimaryTimeout);
1223         futureObj.onComplete(new OnComplete<Object>() {
1224             @Override
1225             public void onComplete(final Throwable failure, final Object response) {
1226                 if (failure != null) {
1227                     handler.onFailure(failure);
1228                 } else {
1229                     if (response instanceof RemotePrimaryShardFound) {
1230                         handler.onRemotePrimaryShardFound((RemotePrimaryShardFound) response);
1231                     } else if (response instanceof LocalPrimaryShardFound) {
1232                         handler.onLocalPrimaryFound((LocalPrimaryShardFound) response);
1233                     } else {
1234                         handler.onUnknownResponse(response);
1235                     }
1236                 }
1237             }
1238         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1239     }
1240
1241     /**
1242      * Construct the name of the shard actor given the name of the member on
1243      * which the shard resides and the name of the shard.
1244      *
1245      * @param memberName the member name
1246      * @param shardName the shard name
1247      * @return a b
1248      */
1249     private ShardIdentifier getShardIdentifier(final MemberName memberName, final String shardName) {
1250         return peerAddressResolver.getShardIdentifier(memberName, shardName);
1251     }
1252
1253     /**
1254      * Create shards that are local to the member on which the ShardManager runs.
1255      */
1256     private void createLocalShards() {
1257         MemberName memberName = this.cluster.getCurrentMemberName();
1258         Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
1259
1260         Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots = new HashMap<>();
1261         if (restoreFromSnapshot != null) {
1262             for (DatastoreSnapshot.ShardSnapshot snapshot: restoreFromSnapshot.getShardSnapshots()) {
1263                 shardSnapshots.put(snapshot.getName(), snapshot);
1264             }
1265         }
1266
1267         // null out to GC
1268         restoreFromSnapshot = null;
1269
1270         for (String shardName : memberShardNames) {
1271             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1272
1273             LOG.debug("{}: Creating local shard: {}", persistenceId(), shardId);
1274
1275             Map<String, String> peerAddresses = getPeerAddresses(shardName);
1276             localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses,
1277                     newShardDatastoreContext(shardName), Shard.builder().restoreFromSnapshot(
1278                         shardSnapshots.get(shardName)), peerAddressResolver));
1279         }
1280     }
1281
1282     /**
1283      * Given the name of the shard find the addresses of all it's peers.
1284      *
1285      * @param shardName the shard name
1286      */
1287     private Map<String, String> getPeerAddresses(final String shardName) {
1288         final Collection<MemberName> members = configuration.getMembersFromShardName(shardName);
1289         return getPeerAddresses(shardName, members);
1290     }
1291
1292     private Map<String, String> getPeerAddresses(final String shardName, final Collection<MemberName> members) {
1293         Map<String, String> peerAddresses = new HashMap<>();
1294         MemberName currentMemberName = this.cluster.getCurrentMemberName();
1295
1296         for (MemberName memberName : members) {
1297             if (!currentMemberName.equals(memberName)) {
1298                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1299                 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
1300                 peerAddresses.put(shardId.toString(), address);
1301             }
1302         }
1303         return peerAddresses;
1304     }
1305
1306     @Override
1307     public SupervisorStrategy supervisorStrategy() {
1308
1309         return new OneForOneStrategy(10, Duration.create("1 minute"),
1310                 (Function<Throwable, Directive>) t -> {
1311                     LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
1312                     return SupervisorStrategy.resume();
1313                 });
1314     }
1315
1316     @Override
1317     public String persistenceId() {
1318         return persistenceId;
1319     }
1320
1321     @VisibleForTesting
1322     ShardManagerInfoMBean getMBean() {
1323         return shardManagerMBean;
1324     }
1325
1326     private boolean isShardReplicaOperationInProgress(final String shardName, final ActorRef sender) {
1327         if (shardReplicaOperationsInProgress.contains(shardName)) {
1328             String msg = String.format("A shard replica operation for %s is already in progress", shardName);
1329             LOG.debug("{}: {}", persistenceId(), msg);
1330             sender.tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1331             return true;
1332         }
1333
1334         return false;
1335     }
1336
1337     private void onAddPrefixShardReplica(final AddPrefixShardReplica message) {
1338         LOG.debug("{}: onAddPrefixShardReplica: {}", persistenceId(), message);
1339
1340         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1341                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1342         final String shardName = shardId.getShardName();
1343
1344         // Create the localShard
1345         if (schemaContext == null) {
1346             String msg = String.format(
1347                     "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1348             LOG.debug("{}: {}", persistenceId(), msg);
1349             getSender().tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1350             return;
1351         }
1352
1353         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1354                 getSelf()) {
1355             @Override
1356             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1357                 final RunnableMessage runnable = (RunnableMessage) () -> addPrefixShard(getShardName(),
1358                         message.getShardPrefix(), response, getSender());
1359                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1360                     getSelf().tell(runnable, getTargetActor());
1361                 }
1362             }
1363
1364             @Override
1365             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1366                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1367             }
1368         });
1369     }
1370
1371     private void onAddShardReplica(final AddShardReplica shardReplicaMsg) {
1372         final String shardName = shardReplicaMsg.getShardName();
1373
1374         LOG.debug("{}: onAddShardReplica: {}", persistenceId(), shardReplicaMsg);
1375
1376         // verify the shard with the specified name is present in the cluster configuration
1377         if (!this.configuration.isShardConfigured(shardName)) {
1378             String msg = String.format("No module configuration exists for shard %s", shardName);
1379             LOG.debug("{}: {}", persistenceId(), msg);
1380             getSender().tell(new Status.Failure(new IllegalArgumentException(msg)), getSelf());
1381             return;
1382         }
1383
1384         // Create the localShard
1385         if (schemaContext == null) {
1386             String msg = String.format(
1387                   "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1388             LOG.debug("{}: {}", persistenceId(), msg);
1389             getSender().tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1390             return;
1391         }
1392
1393         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(),
1394                 getSelf()) {
1395             @Override
1396             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1397                 final RunnableMessage runnable = (RunnableMessage) () ->
1398                     addShard(getShardName(), response, getSender());
1399                 if (!isPreviousShardActorStopInProgress(getShardName(), runnable)) {
1400                     getSelf().tell(runnable, getTargetActor());
1401                 }
1402             }
1403
1404             @Override
1405             public void onLocalPrimaryFound(final LocalPrimaryShardFound message) {
1406                 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1407             }
1408         });
1409     }
1410
1411     private void sendLocalReplicaAlreadyExistsReply(final String shardName, final ActorRef sender) {
1412         String msg = String.format("Local shard %s already exists", shardName);
1413         LOG.debug("{}: {}", persistenceId(), msg);
1414         sender.tell(new Status.Failure(new AlreadyExistsException(msg)), getSelf());
1415     }
1416
1417     private void addPrefixShard(final String shardName, final YangInstanceIdentifier shardPrefix,
1418                                 final RemotePrimaryShardFound response, final ActorRef sender) {
1419         if (isShardReplicaOperationInProgress(shardName, sender)) {
1420             return;
1421         }
1422
1423         shardReplicaOperationsInProgress.add(shardName);
1424
1425         final ShardInformation shardInfo;
1426         final boolean removeShardOnFailure;
1427         ShardInformation existingShardInfo = localShards.get(shardName);
1428         if (existingShardInfo == null) {
1429             removeShardOnFailure = true;
1430             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1431
1432             final Builder builder = newShardDatastoreContextBuilder(shardName);
1433             builder.storeRoot(shardPrefix).customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName());
1434
1435             DatastoreContext datastoreContext = builder.build();
1436
1437             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1438                     Shard.builder(), peerAddressResolver);
1439             shardInfo.setActiveMember(false);
1440             shardInfo.setSchemaContext(schemaContext);
1441             localShards.put(shardName, shardInfo);
1442             shardInfo.setActor(newShardActor(shardInfo));
1443         } else {
1444             removeShardOnFailure = false;
1445             shardInfo = existingShardInfo;
1446         }
1447
1448         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1449     }
1450
1451     private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) {
1452         if (isShardReplicaOperationInProgress(shardName, sender)) {
1453             return;
1454         }
1455
1456         shardReplicaOperationsInProgress.add(shardName);
1457
1458         final ShardInformation shardInfo;
1459         final boolean removeShardOnFailure;
1460         ShardInformation existingShardInfo = localShards.get(shardName);
1461         if (existingShardInfo == null) {
1462             removeShardOnFailure = true;
1463             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1464
1465             DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName)
1466                     .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
1467
1468             shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1469                     Shard.builder(), peerAddressResolver);
1470             shardInfo.setActiveMember(false);
1471             shardInfo.setSchemaContext(schemaContext);
1472             localShards.put(shardName, shardInfo);
1473             shardInfo.setActor(newShardActor(shardInfo));
1474         } else {
1475             removeShardOnFailure = false;
1476             shardInfo = existingShardInfo;
1477         }
1478
1479         execAddShard(shardName, shardInfo, response, removeShardOnFailure, sender);
1480     }
1481
1482     private void execAddShard(final String shardName,
1483                               final ShardInformation shardInfo,
1484                               final RemotePrimaryShardFound response,
1485                               final boolean removeShardOnFailure,
1486                               final ActorRef sender) {
1487
1488         final String localShardAddress =
1489                 peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName());
1490
1491         //inform ShardLeader to add this shard as a replica by sending an AddServer message
1492         LOG.debug("{}: Sending AddServer message to peer {} for shard {}", persistenceId(),
1493                 response.getPrimaryPath(), shardInfo.getShardId());
1494
1495         final Timeout addServerTimeout = new Timeout(shardInfo.getDatastoreContext()
1496                 .getShardLeaderElectionTimeout().duration());
1497         final Future<Object> futureObj = ask(getContext().actorSelection(response.getPrimaryPath()),
1498                 new AddServer(shardInfo.getShardId().toString(), localShardAddress, true), addServerTimeout);
1499
1500         futureObj.onComplete(new OnComplete<Object>() {
1501             @Override
1502             public void onComplete(final Throwable failure, final Object addServerResponse) {
1503                 if (failure != null) {
1504                     LOG.debug("{}: AddServer request to {} for {} failed", persistenceId(),
1505                             response.getPrimaryPath(), shardName, failure);
1506
1507                     final String msg = String.format("AddServer request to leader %s for shard %s failed",
1508                             response.getPrimaryPath(), shardName);
1509                     self().tell(new ForwardedAddServerFailure(shardName, msg, failure, removeShardOnFailure), sender);
1510                 } else {
1511                     self().tell(new ForwardedAddServerReply(shardInfo, (AddServerReply)addServerResponse,
1512                             response.getPrimaryPath(), removeShardOnFailure), sender);
1513                 }
1514             }
1515         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1516     }
1517
1518     private void onAddServerFailure(final String shardName, final String message, final Throwable failure,
1519             final ActorRef sender, final boolean removeShardOnFailure) {
1520         shardReplicaOperationsInProgress.remove(shardName);
1521
1522         if (removeShardOnFailure) {
1523             ShardInformation shardInfo = localShards.remove(shardName);
1524             if (shardInfo.getActor() != null) {
1525                 shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf());
1526             }
1527         }
1528
1529         sender.tell(new Status.Failure(message == null ? failure :
1530             new RuntimeException(message, failure)), getSelf());
1531     }
1532
1533     private void onAddServerReply(final ShardInformation shardInfo, final AddServerReply replyMsg,
1534             final ActorRef sender, final String leaderPath, final boolean removeShardOnFailure) {
1535         String shardName = shardInfo.getShardName();
1536         shardReplicaOperationsInProgress.remove(shardName);
1537
1538         LOG.debug("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath);
1539
1540         if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1541             LOG.debug("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName);
1542
1543             // Make the local shard voting capable
1544             shardInfo.setDatastoreContext(newShardDatastoreContext(shardName), getSelf());
1545             shardInfo.setActiveMember(true);
1546             persistShardList();
1547
1548             sender.tell(new Status.Success(null), getSelf());
1549         } else if (replyMsg.getStatus() == ServerChangeStatus.ALREADY_EXISTS) {
1550             sendLocalReplicaAlreadyExistsReply(shardName, sender);
1551         } else {
1552             LOG.warn("{}: Leader failed to add shard replica {} with status {}",
1553                     persistenceId(), shardName, replyMsg.getStatus());
1554
1555             Exception failure = getServerChangeException(AddServer.class, replyMsg.getStatus(), leaderPath,
1556                     shardInfo.getShardId());
1557
1558             onAddServerFailure(shardName, null, failure, sender, removeShardOnFailure);
1559         }
1560     }
1561
1562     private static Exception getServerChangeException(final Class<?> serverChange,
1563             final ServerChangeStatus serverChangeStatus, final String leaderPath, final ShardIdentifier shardId) {
1564         switch (serverChangeStatus) {
1565             case TIMEOUT:
1566                 return new TimeoutException(String.format(
1567                         "The shard leader %s timed out trying to replicate the initial data to the new shard %s."
1568                         + "Possible causes - there was a problem replicating the data or shard leadership changed "
1569                         + "while replicating the shard data", leaderPath, shardId.getShardName()));
1570             case NO_LEADER:
1571                 return createNoShardLeaderException(shardId);
1572             case NOT_SUPPORTED:
1573                 return new UnsupportedOperationException(String.format("%s request is not supported for shard %s",
1574                         serverChange.getSimpleName(), shardId.getShardName()));
1575             default :
1576                 return new RuntimeException(String.format("%s request to leader %s for shard %s failed with status %s",
1577                         serverChange.getSimpleName(), leaderPath, shardId.getShardName(), serverChangeStatus));
1578         }
1579     }
1580
1581     private void onRemoveShardReplica(final RemoveShardReplica shardReplicaMsg) {
1582         LOG.debug("{}: onRemoveShardReplica: {}", persistenceId(), shardReplicaMsg);
1583
1584         findPrimary(shardReplicaMsg.getShardName(), new AutoFindPrimaryFailureResponseHandler(getSender(),
1585                 shardReplicaMsg.getShardName(), persistenceId(), getSelf()) {
1586             @Override
1587             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1588                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1589             }
1590
1591             @Override
1592             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1593                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1594             }
1595
1596             private void doRemoveShardReplicaAsync(final String primaryPath) {
1597                 getSelf().tell((RunnableMessage) () -> removeShardReplica(shardReplicaMsg, getShardName(),
1598                         primaryPath, getSender()), getTargetActor());
1599             }
1600         });
1601     }
1602
1603     private void onRemovePrefixShardReplica(final RemovePrefixShardReplica message) {
1604         LOG.debug("{}: onRemovePrefixShardReplica: {}", persistenceId(), message);
1605
1606         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(),
1607                 ClusterUtils.getCleanShardName(message.getShardPrefix()));
1608         final String shardName = shardId.getShardName();
1609
1610         findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(),
1611                 shardName, persistenceId(), getSelf()) {
1612             @Override
1613             public void onRemotePrimaryShardFound(final RemotePrimaryShardFound response) {
1614                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1615             }
1616
1617             @Override
1618             public void onLocalPrimaryFound(final LocalPrimaryShardFound response) {
1619                 doRemoveShardReplicaAsync(response.getPrimaryPath());
1620             }
1621
1622             private void doRemoveShardReplicaAsync(final String primaryPath) {
1623                 getSelf().tell((RunnableMessage) () -> removePrefixShardReplica(message, getShardName(),
1624                         primaryPath, getSender()), getTargetActor());
1625             }
1626         });
1627     }
1628
1629     private void persistShardList() {
1630         List<String> shardList = new ArrayList<>(localShards.keySet());
1631         for (ShardInformation shardInfo : localShards.values()) {
1632             if (!shardInfo.isActiveMember()) {
1633                 shardList.remove(shardInfo.getShardName());
1634             }
1635         }
1636         LOG.debug("{}: persisting the shard list {}", persistenceId(), shardList);
1637         saveSnapshot(updateShardManagerSnapshot(shardList, configuration.getAllPrefixShardConfigurations()));
1638     }
1639
1640     private ShardManagerSnapshot updateShardManagerSnapshot(
1641             final List<String> shardList,
1642             final Map<DOMDataTreeIdentifier, PrefixShardConfiguration> allPrefixShardConfigurations) {
1643         currentSnapshot = new ShardManagerSnapshot(shardList, allPrefixShardConfigurations);
1644         return currentSnapshot;
1645     }
1646
1647     private void applyShardManagerSnapshot(final ShardManagerSnapshot snapshot) {
1648         currentSnapshot = snapshot;
1649
1650         LOG.debug("{}: onSnapshotOffer: {}", persistenceId(), currentSnapshot);
1651
1652         final MemberName currentMember = cluster.getCurrentMemberName();
1653         Set<String> configuredShardList =
1654             new HashSet<>(configuration.getMemberShardNames(currentMember));
1655         for (String shard : currentSnapshot.getShardList()) {
1656             if (!configuredShardList.contains(shard)) {
1657                 // add the current member as a replica for the shard
1658                 LOG.debug("{}: adding shard {}", persistenceId(), shard);
1659                 configuration.addMemberReplicaForShard(shard, currentMember);
1660             } else {
1661                 configuredShardList.remove(shard);
1662             }
1663         }
1664         for (String shard : configuredShardList) {
1665             // remove the member as a replica for the shard
1666             LOG.debug("{}: removing shard {}", persistenceId(), shard);
1667             configuration.removeMemberReplicaForShard(shard, currentMember);
1668         }
1669     }
1670
1671     private void onSaveSnapshotSuccess(final SaveSnapshotSuccess successMessage) {
1672         LOG.debug("{} saved ShardManager snapshot successfully. Deleting the prev snapshot if available",
1673             persistenceId());
1674         deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), successMessage.metadata().timestamp() - 1,
1675             0, 0));
1676     }
1677
1678     private void onChangeShardServersVotingStatus(final ChangeShardMembersVotingStatus changeMembersVotingStatus) {
1679         LOG.debug("{}: onChangeShardServersVotingStatus: {}", persistenceId(), changeMembersVotingStatus);
1680
1681         String shardName = changeMembersVotingStatus.getShardName();
1682         Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1683         for (Entry<String, Boolean> e: changeMembersVotingStatus.getMeberVotingStatusMap().entrySet()) {
1684             serverVotingStatusMap.put(getShardIdentifier(MemberName.forName(e.getKey()), shardName).toString(),
1685                     e.getValue());
1686         }
1687
1688         ChangeServersVotingStatus changeServersVotingStatus = new ChangeServersVotingStatus(serverVotingStatusMap);
1689
1690         findLocalShard(shardName, getSender(),
1691             localShardFound -> changeShardMembersVotingStatus(changeServersVotingStatus, shardName,
1692             localShardFound.getPath(), getSender()));
1693     }
1694
1695     private void onFlipShardMembersVotingStatus(final FlipShardMembersVotingStatus flipMembersVotingStatus) {
1696         LOG.debug("{}: onFlipShardMembersVotingStatus: {}", persistenceId(), flipMembersVotingStatus);
1697
1698         ActorRef sender = getSender();
1699         final String shardName = flipMembersVotingStatus.getShardName();
1700         findLocalShard(shardName, sender, localShardFound -> {
1701             Future<Object> future = ask(localShardFound.getPath(), GetOnDemandRaftState.INSTANCE,
1702                     Timeout.apply(30, TimeUnit.SECONDS));
1703
1704             future.onComplete(new OnComplete<Object>() {
1705                 @Override
1706                 public void onComplete(final Throwable failure, final Object response) {
1707                     if (failure != null) {
1708                         sender.tell(new Status.Failure(new RuntimeException(
1709                                 String.format("Failed to access local shard %s", shardName), failure)), self());
1710                         return;
1711                     }
1712
1713                     OnDemandRaftState raftState = (OnDemandRaftState) response;
1714                     Map<String, Boolean> serverVotingStatusMap = new HashMap<>();
1715                     for (Entry<String, Boolean> e: raftState.getPeerVotingStates().entrySet()) {
1716                         serverVotingStatusMap.put(e.getKey(), !e.getValue());
1717                     }
1718
1719                     serverVotingStatusMap.put(getShardIdentifier(cluster.getCurrentMemberName(), shardName)
1720                             .toString(), !raftState.isVoting());
1721
1722                     changeShardMembersVotingStatus(new ChangeServersVotingStatus(serverVotingStatusMap),
1723                             shardName, localShardFound.getPath(), sender);
1724                 }
1725             }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1726         });
1727
1728     }
1729
1730     private void findLocalShard(final FindLocalShard message) {
1731         LOG.debug("{}: findLocalShard : {}", persistenceId(), message.getShardName());
1732
1733         final ShardInformation shardInformation = localShards.get(message.getShardName());
1734
1735         if (shardInformation == null) {
1736             LOG.debug("{}: Local shard {} not found - shards present: {}",
1737                     persistenceId(), message.getShardName(), localShards.keySet());
1738
1739             getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
1740             return;
1741         }
1742
1743         sendResponse(shardInformation, message.isWaitUntilInitialized(), false,
1744             () -> new LocalShardFound(shardInformation.getActor()));
1745     }
1746
1747     private void findLocalShard(final String shardName, final ActorRef sender,
1748             final Consumer<LocalShardFound> onLocalShardFound) {
1749         Timeout findLocalTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext()
1750                 .getShardInitializationTimeout().duration().$times(2));
1751
1752         Future<Object> futureObj = ask(getSelf(), new FindLocalShard(shardName, true), findLocalTimeout);
1753         futureObj.onComplete(new OnComplete<Object>() {
1754             @Override
1755             public void onComplete(final Throwable failure, final Object response) {
1756                 if (failure != null) {
1757                     LOG.debug("{}: Received failure from FindLocalShard for shard {}", persistenceId, shardName,
1758                             failure);
1759                     sender.tell(new Status.Failure(new RuntimeException(
1760                             String.format("Failed to find local shard %s", shardName), failure)), self());
1761                 } else {
1762                     if (response instanceof LocalShardFound) {
1763                         getSelf().tell((RunnableMessage) () -> onLocalShardFound.accept((LocalShardFound) response),
1764                                 sender);
1765                     } else if (response instanceof LocalShardNotFound) {
1766                         String msg = String.format("Local shard %s does not exist", shardName);
1767                         LOG.debug("{}: {}", persistenceId, msg);
1768                         sender.tell(new Status.Failure(new IllegalArgumentException(msg)), self());
1769                     } else {
1770                         String msg = String.format("Failed to find local shard %s: received response: %s",
1771                                 shardName, response);
1772                         LOG.debug("{}: {}", persistenceId, msg);
1773                         sender.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response :
1774                                 new RuntimeException(msg)), self());
1775                     }
1776                 }
1777             }
1778         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1779     }
1780
1781     private void changeShardMembersVotingStatus(final ChangeServersVotingStatus changeServersVotingStatus,
1782             final String shardName, final ActorRef shardActorRef, final ActorRef sender) {
1783         if (isShardReplicaOperationInProgress(shardName, sender)) {
1784             return;
1785         }
1786
1787         shardReplicaOperationsInProgress.add(shardName);
1788
1789         DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
1790         final ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1791
1792         LOG.debug("{}: Sending ChangeServersVotingStatus message {} to local shard {}", persistenceId(),
1793                 changeServersVotingStatus, shardActorRef.path());
1794
1795         Timeout timeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().duration().$times(2));
1796         Future<Object> futureObj = ask(shardActorRef, changeServersVotingStatus, timeout);
1797
1798         futureObj.onComplete(new OnComplete<Object>() {
1799             @Override
1800             public void onComplete(final Throwable failure, final Object response) {
1801                 shardReplicaOperationsInProgress.remove(shardName);
1802                 if (failure != null) {
1803                     String msg = String.format("ChangeServersVotingStatus request to local shard %s failed",
1804                             shardActorRef.path());
1805                     LOG.debug("{}: {}", persistenceId(), msg, failure);
1806                     sender.tell(new Status.Failure(new RuntimeException(msg, failure)), self());
1807                 } else {
1808                     LOG.debug("{}: Received {} from local shard {}", persistenceId(), response, shardActorRef.path());
1809
1810                     ServerChangeReply replyMsg = (ServerChangeReply) response;
1811                     if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1812                         LOG.debug("{}: ChangeServersVotingStatus succeeded for shard {}", persistenceId(), shardName);
1813                         sender.tell(new Status.Success(null), getSelf());
1814                     } else if (replyMsg.getStatus() == ServerChangeStatus.INVALID_REQUEST) {
1815                         sender.tell(new Status.Failure(new IllegalArgumentException(String.format(
1816                                 "The requested voting state change for shard %s is invalid. At least one member "
1817                                 + "must be voting", shardId.getShardName()))), getSelf());
1818                     } else {
1819                         LOG.warn("{}: ChangeServersVotingStatus failed for shard {} with status {}",
1820                                 persistenceId(), shardName, replyMsg.getStatus());
1821
1822                         Exception error = getServerChangeException(ChangeServersVotingStatus.class,
1823                                 replyMsg.getStatus(), shardActorRef.path().toString(), shardId);
1824                         sender.tell(new Status.Failure(error), getSelf());
1825                     }
1826                 }
1827             }
1828         }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1829     }
1830
1831     private static final class ForwardedAddServerReply {
1832         ShardInformation shardInfo;
1833         AddServerReply addServerReply;
1834         String leaderPath;
1835         boolean removeShardOnFailure;
1836
1837         ForwardedAddServerReply(final ShardInformation shardInfo, final AddServerReply addServerReply,
1838             final String leaderPath, final boolean removeShardOnFailure) {
1839             this.shardInfo = shardInfo;
1840             this.addServerReply = addServerReply;
1841             this.leaderPath = leaderPath;
1842             this.removeShardOnFailure = removeShardOnFailure;
1843         }
1844     }
1845
1846     private static final class ForwardedAddServerFailure {
1847         String shardName;
1848         String failureMessage;
1849         Throwable failure;
1850         boolean removeShardOnFailure;
1851
1852         ForwardedAddServerFailure(final String shardName, final String failureMessage, final Throwable failure,
1853                 final boolean removeShardOnFailure) {
1854             this.shardName = shardName;
1855             this.failureMessage = failureMessage;
1856             this.failure = failure;
1857             this.removeShardOnFailure = removeShardOnFailure;
1858         }
1859     }
1860
1861     static class OnShardInitialized {
1862         private final Runnable replyRunnable;
1863         private Cancellable timeoutSchedule;
1864
1865         OnShardInitialized(final Runnable replyRunnable) {
1866             this.replyRunnable = replyRunnable;
1867         }
1868
1869         Runnable getReplyRunnable() {
1870             return replyRunnable;
1871         }
1872
1873         Cancellable getTimeoutSchedule() {
1874             return timeoutSchedule;
1875         }
1876
1877         void setTimeoutSchedule(final Cancellable timeoutSchedule) {
1878             this.timeoutSchedule = timeoutSchedule;
1879         }
1880     }
1881
1882     static class OnShardReady extends OnShardInitialized {
1883         OnShardReady(final Runnable replyRunnable) {
1884             super(replyRunnable);
1885         }
1886     }
1887
1888     private interface RunnableMessage extends Runnable {
1889     }
1890
1891     /**
1892      * The FindPrimaryResponseHandler provides specific callback methods which are invoked when a response to the
1893      * a remote or local find primary message is processed.
1894      */
1895     private interface FindPrimaryResponseHandler {
1896         /**
1897          * Invoked when a Failure message is received as a response.
1898          *
1899          * @param failure the failure exception
1900          */
1901         void onFailure(Throwable failure);
1902
1903         /**
1904          * Invoked when a RemotePrimaryShardFound response is received.
1905          *
1906          * @param response the response
1907          */
1908         void onRemotePrimaryShardFound(RemotePrimaryShardFound response);
1909
1910         /**
1911          * Invoked when a LocalPrimaryShardFound response is received.
1912          *
1913          * @param response the response
1914          */
1915         void onLocalPrimaryFound(LocalPrimaryShardFound response);
1916
1917         /**
1918          * Invoked when an unknown response is received. This is another type of failure.
1919          *
1920          * @param response the response
1921          */
1922         void onUnknownResponse(Object response);
1923     }
1924
1925     /**
1926      * The AutoFindPrimaryFailureResponseHandler automatically processes Failure responses when finding a primary
1927      * replica and sends a wrapped Failure response to some targetActor.
1928      */
1929     private abstract static class AutoFindPrimaryFailureResponseHandler implements FindPrimaryResponseHandler {
1930         private final ActorRef targetActor;
1931         private final String shardName;
1932         private final String persistenceId;
1933         private final ActorRef shardManagerActor;
1934
1935         /**
1936          * Constructs an instance.
1937          *
1938          * @param targetActor The actor to whom the Failure response should be sent when a FindPrimary failure occurs
1939          * @param shardName The name of the shard for which the primary replica had to be found
1940          * @param persistenceId The persistenceId for the ShardManager
1941          * @param shardManagerActor The ShardManager actor which triggered the call to FindPrimary
1942          */
1943         protected AutoFindPrimaryFailureResponseHandler(final ActorRef targetActor, final String shardName,
1944                 final String persistenceId, final ActorRef shardManagerActor) {
1945             this.targetActor = Preconditions.checkNotNull(targetActor);
1946             this.shardName = Preconditions.checkNotNull(shardName);
1947             this.persistenceId = Preconditions.checkNotNull(persistenceId);
1948             this.shardManagerActor = Preconditions.checkNotNull(shardManagerActor);
1949         }
1950
1951         public ActorRef getTargetActor() {
1952             return targetActor;
1953         }
1954
1955         public String getShardName() {
1956             return shardName;
1957         }
1958
1959         @Override
1960         public void onFailure(final Throwable failure) {
1961             LOG.debug("{}: Received failure from FindPrimary for shard {}", persistenceId, shardName, failure);
1962             targetActor.tell(new Status.Failure(new RuntimeException(
1963                     String.format("Failed to find leader for shard %s", shardName), failure)), shardManagerActor);
1964         }
1965
1966         @Override
1967         public void onUnknownResponse(final Object response) {
1968             String msg = String.format("Failed to find leader for shard %s: received response: %s",
1969                     shardName, response);
1970             LOG.debug("{}: {}", persistenceId, msg);
1971             targetActor.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response :
1972                     new RuntimeException(msg)), shardManagerActor);
1973         }
1974     }
1975
1976     /**
1977      * The WrappedShardResponse class wraps a response from a Shard.
1978      */
1979     private static final class WrappedShardResponse {
1980         private final ShardIdentifier shardId;
1981         private final Object response;
1982         private final String leaderPath;
1983
1984         WrappedShardResponse(final ShardIdentifier shardId, final Object response, final String leaderPath) {
1985             this.shardId = shardId;
1986             this.response = response;
1987             this.leaderPath = leaderPath;
1988         }
1989
1990         ShardIdentifier getShardId() {
1991             return shardId;
1992         }
1993
1994         Object getResponse() {
1995             return response;
1996         }
1997
1998         String getLeaderPath() {
1999             return leaderPath;
2000         }
2001     }
2002
2003     private static final class ShardNotInitializedTimeout {
2004         private final ActorRef sender;
2005         private final ShardInformation shardInfo;
2006         private final OnShardInitialized onShardInitialized;
2007
2008         ShardNotInitializedTimeout(final ShardInformation shardInfo, final OnShardInitialized onShardInitialized,
2009             final ActorRef sender) {
2010             this.sender = sender;
2011             this.shardInfo = shardInfo;
2012             this.onShardInitialized = onShardInitialized;
2013         }
2014
2015         ActorRef getSender() {
2016             return sender;
2017         }
2018
2019         ShardInformation getShardInfo() {
2020             return shardInfo;
2021         }
2022
2023         OnShardInitialized getOnShardInitialized() {
2024             return onShardInitialized;
2025         }
2026     }
2027 }
2028
2029
2030