opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/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;
  10
  11 import akka.actor.ActorPath;
  12 import akka.actor.ActorRef;
  13 import akka.actor.Address;
  14 import akka.actor.Cancellable;
  15 import akka.actor.OneForOneStrategy;
  16 import akka.actor.Props;
  17 import akka.actor.SupervisorStrategy;
  18 import akka.cluster.ClusterEvent;
  19 import akka.japi.Creator;
  20 import akka.japi.Function;
  21 import akka.persistence.RecoveryCompleted;
  22 import akka.serialization.Serialization;
  23 import com.google.common.annotations.VisibleForTesting;
  24 import com.google.common.base.Objects;
  25 import com.google.common.base.Optional;
  26 import com.google.common.base.Preconditions;
  27 import com.google.common.base.Strings;
  28 import com.google.common.base.Supplier;
  29 import com.google.common.collect.Sets;
  30 import java.io.Serializable;
  31 import java.util.ArrayList;
  32 import java.util.Collection;
  33 import java.util.HashMap;
  34 import java.util.Iterator;
  35 import java.util.List;
  36 import java.util.Map;
  37 import java.util.Set;
  38 import java.util.concurrent.CountDownLatch;
  39 import java.util.concurrent.TimeUnit;
  40 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActorWithMetering;
  41 import org.opendaylight.controller.cluster.datastore.config.Configuration;
  42 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  43 import org.opendaylight.controller.cluster.datastore.exceptions.NotInitializedException;
  44 import org.opendaylight.controller.cluster.datastore.exceptions.PrimaryNotFoundException;
  45 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
  46 import org.opendaylight.controller.cluster.datastore.identifiers.ShardManagerIdentifier;
  47 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfo;
  48 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfoMBean;
  49 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  50 import org.opendaylight.controller.cluster.datastore.messages.CreateShard;
  51 import org.opendaylight.controller.cluster.datastore.messages.CreateShardReply;
  52 import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard;
  53 import org.opendaylight.controller.cluster.datastore.messages.FindPrimary;
  54 import org.opendaylight.controller.cluster.datastore.messages.LocalPrimaryShardFound;
  55 import org.opendaylight.controller.cluster.datastore.messages.LocalShardFound;
  56 import org.opendaylight.controller.cluster.datastore.messages.LocalShardNotFound;
  57 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  58 import org.opendaylight.controller.cluster.datastore.messages.PeerDown;
  59 import org.opendaylight.controller.cluster.datastore.messages.PeerUp;
  60 import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary;
  61 import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound;
  62 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  63 import org.opendaylight.controller.cluster.datastore.messages.SwitchShardBehavior;
  64 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  65 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  66 import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache;
  67 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  68 import org.opendaylight.controller.cluster.notifications.RoleChangeNotification;
  69 import org.opendaylight.controller.cluster.raft.RaftState;
  70 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  71 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  73 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  74 import org.slf4j.Logger;
  75 import org.slf4j.LoggerFactory;
  76 import scala.concurrent.duration.Duration;
  77 import scala.concurrent.duration.FiniteDuration;
  78
  79 /**
  80  * The ShardManager has the following jobs,
  81  * <ul>
  82  * <li> Create all the local shard replicas that belong on this cluster member
  83  * <li> Find the address of the local shard
  84  * <li> Find the primary replica for any given shard
  85  * <li> Monitor the cluster members and store their addresses
  86  * <ul>
  87  */
  88 public class ShardManager extends AbstractUntypedPersistentActorWithMetering {
  89
  90     private static final Logger LOG = LoggerFactory.getLogger(ShardManager.class);
  91
  92     // Stores a mapping between a member name and the address of the member
  93     // Member names look like "member-1", "member-2" etc and are as specified
  94     // in configuration
  95     private final Map<String, Address> memberNameToAddress = new HashMap<>();
  96
  97     // Stores a mapping between a shard name and it's corresponding information
  98     // Shard names look like inventory, topology etc and are as specified in
  99     // configuration
 100     private final Map<String, ShardInformation> localShards = new HashMap<>();
 101
 102     // The type of a ShardManager reflects the type of the datastore itself
 103     // A data store could be of type config/operational
 104     private final String type;
 105
 106     private final String shardManagerIdentifierString;
 107
 108     private final ClusterWrapper cluster;
 109
 110     private final Configuration configuration;
 111
 112     private final String shardDispatcherPath;
 113
 114     private ShardManagerInfo mBean;
 115
 116     private DatastoreContext datastoreContext;
 117
 118     private final CountDownLatch waitTillReadyCountdownLatch;
 119
 120     private final PrimaryShardInfoFutureCache primaryShardInfoCache;
 121
 122     private SchemaContext schemaContext;
 123
 124     /**
 125      */
 126     protected ShardManager(ClusterWrapper cluster, Configuration configuration,
 127             DatastoreContext datastoreContext, CountDownLatch waitTillReadyCountdownLatch,
 128             PrimaryShardInfoFutureCache primaryShardInfoCache) {
 129
 130         this.cluster = Preconditions.checkNotNull(cluster, "cluster should not be null");
 131         this.configuration = Preconditions.checkNotNull(configuration, "configuration should not be null");
 132         this.datastoreContext = datastoreContext;
 133         this.type = datastoreContext.getDataStoreType();
 134         this.shardManagerIdentifierString = ShardManagerIdentifier.builder().type(type).build().toString();
 135         this.shardDispatcherPath =
 136                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Shard);
 137         this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
 138         this.primaryShardInfoCache = primaryShardInfoCache;
 139
 140         // Subscribe this actor to cluster member events
 141         cluster.subscribeToMemberEvents(getSelf());
 142
 143         createLocalShards();
 144     }
 145
 146     public static Props props(
 147         final ClusterWrapper cluster,
 148         final Configuration configuration,
 149         final DatastoreContext datastoreContext,
 150         final CountDownLatch waitTillReadyCountdownLatch,
 151         final PrimaryShardInfoFutureCache primaryShardInfoCache) {
 152
 153         Preconditions.checkNotNull(cluster, "cluster should not be null");
 154         Preconditions.checkNotNull(configuration, "configuration should not be null");
 155         Preconditions.checkNotNull(waitTillReadyCountdownLatch, "waitTillReadyCountdownLatch should not be null");
 156         Preconditions.checkNotNull(primaryShardInfoCache, "primaryShardInfoCache should not be null");
 157
 158         return Props.create(new ShardManagerCreator(cluster, configuration, datastoreContext,
 159                 waitTillReadyCountdownLatch, primaryShardInfoCache));
 160     }
 161
 162     @Override
 163     public void postStop() {
 164         LOG.info("Stopping ShardManager");
 165
 166         mBean.unregisterMBean();
 167     }
 168
 169     @Override
 170     public void handleCommand(Object message) throws Exception {
 171         if (message  instanceof FindPrimary) {
 172             findPrimary((FindPrimary)message);
 173         } else if(message instanceof FindLocalShard){
 174             findLocalShard((FindLocalShard) message);
 175         } else if (message instanceof UpdateSchemaContext) {
 176             updateSchemaContext(message);
 177         } else if(message instanceof ActorInitialized) {
 178             onActorInitialized(message);
 179         } else if (message instanceof ClusterEvent.MemberUp){
 180             memberUp((ClusterEvent.MemberUp) message);
 181         } else if (message instanceof ClusterEvent.MemberExited){
 182             memberExited((ClusterEvent.MemberExited) message);
 183         } else if(message instanceof ClusterEvent.MemberRemoved) {
 184             memberRemoved((ClusterEvent.MemberRemoved) message);
 185         } else if(message instanceof ClusterEvent.UnreachableMember) {
 186             memberUnreachable((ClusterEvent.UnreachableMember)message);
 187         } else if(message instanceof ClusterEvent.ReachableMember) {
 188             memberReachable((ClusterEvent.ReachableMember) message);
 189         } else if(message instanceof DatastoreContext) {
 190             onDatastoreContext((DatastoreContext)message);
 191         } else if(message instanceof RoleChangeNotification) {
 192             onRoleChangeNotification((RoleChangeNotification) message);
 193         } else if(message instanceof FollowerInitialSyncUpStatus){
 194             onFollowerInitialSyncStatus((FollowerInitialSyncUpStatus) message);
 195         } else if(message instanceof ShardNotInitializedTimeout) {
 196             onShardNotInitializedTimeout((ShardNotInitializedTimeout)message);
 197         } else if(message instanceof ShardLeaderStateChanged) {
 198             onLeaderStateChanged((ShardLeaderStateChanged) message);
 199         } else if(message instanceof SwitchShardBehavior){
 200             onSwitchShardBehavior((SwitchShardBehavior) message);
 201         } else if(message instanceof CreateShard) {
 202             onCreateShard((CreateShard)message);
 203         } else {
 204             unknownMessage(message);
 205         }
 206
 207     }
 208
 209     private void onCreateShard(CreateShard createShard) {
 210         Object reply;
 211         try {
 212             if(localShards.containsKey(createShard.getShardName())) {
 213                 throw new IllegalStateException(String.format("Shard with name %s already exists",
 214                         createShard.getShardName()));
 215             }
 216
 217             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), createShard.getShardName());
 218             Map<String, String> peerAddresses = getPeerAddresses(createShard.getShardName(), createShard.getMemberNames());
 219
 220             LOG.debug("onCreateShard: shardId: {}, memberNames: {}. peerAddresses: {}", shardId,
 221                     createShard.getMemberNames(), peerAddresses);
 222
 223             DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
 224             if(shardDatastoreContext == null) {
 225                 shardDatastoreContext = datastoreContext;
 226             }
 227
 228             ShardInformation info = new ShardInformation(createShard.getShardName(), shardId, peerAddresses,
 229                     shardDatastoreContext, createShard.getShardPropsCreator());
 230             localShards.put(createShard.getShardName(), info);
 231
 232             mBean.addLocalShard(shardId.toString());
 233
 234             if(schemaContext != null) {
 235                 info.setActor(newShardActor(schemaContext, info));
 236             }
 237
 238             reply = new CreateShardReply();
 239         } catch (Exception e) {
 240             LOG.error("onCreateShard failed", e);
 241             reply = new akka.actor.Status.Failure(e);
 242         }
 243
 244         if(getSender() != null && !getContext().system().deadLetters().equals(getSender())) {
 245             getSender().tell(reply, getSelf());
 246         }
 247     }
 248
 249     private void checkReady(){
 250         if (isReadyWithLeaderId()) {
 251             LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
 252                     persistenceId(), type, waitTillReadyCountdownLatch.getCount());
 253
 254             waitTillReadyCountdownLatch.countDown();
 255         }
 256     }
 257
 258     private void onLeaderStateChanged(ShardLeaderStateChanged leaderStateChanged) {
 259         LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
 260
 261         ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
 262         if(shardInformation != null) {
 263             shardInformation.setLocalDataTree(leaderStateChanged.getLocalShardDataTree());
 264             shardInformation.setLeaderVersion(leaderStateChanged.getLeaderPayloadVersion());
 265             if(shardInformation.setLeaderId(leaderStateChanged.getLeaderId())) {
 266                 primaryShardInfoCache.remove(shardInformation.getShardName());
 267             }
 268
 269             checkReady();
 270         } else {
 271             LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
 272         }
 273     }
 274
 275     private void onShardNotInitializedTimeout(ShardNotInitializedTimeout message) {
 276         ShardInformation shardInfo = message.getShardInfo();
 277
 278         LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
 279                 shardInfo.getShardName());
 280
 281         shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
 282
 283         if(!shardInfo.isShardInitialized()) {
 284             LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
 285             message.getSender().tell(createNotInitializedException(shardInfo.shardId), getSelf());
 286         } else {
 287             LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
 288             message.getSender().tell(createNoShardLeaderException(shardInfo.shardId), getSelf());
 289         }
 290     }
 291
 292     private void onFollowerInitialSyncStatus(FollowerInitialSyncUpStatus status) {
 293         LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
 294                 status.getName(), status.isInitialSyncDone());
 295
 296         ShardInformation shardInformation = findShardInformation(status.getName());
 297
 298         if(shardInformation != null) {
 299             shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
 300
 301             mBean.setSyncStatus(isInSync());
 302         }
 303
 304     }
 305
 306     private void onRoleChangeNotification(RoleChangeNotification roleChanged) {
 307         LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
 308                 roleChanged.getOldRole(), roleChanged.getNewRole());
 309
 310         ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
 311         if(shardInformation != null) {
 312             shardInformation.setRole(roleChanged.getNewRole());
 313             checkReady();
 314             mBean.setSyncStatus(isInSync());
 315         }
 316     }
 317
 318
 319     private ShardInformation findShardInformation(String memberId) {
 320         for(ShardInformation info : localShards.values()){
 321             if(info.getShardId().toString().equals(memberId)){
 322                 return info;
 323             }
 324         }
 325
 326         return null;
 327     }
 328
 329     private boolean isReadyWithLeaderId() {
 330         boolean isReady = true;
 331         for (ShardInformation info : localShards.values()) {
 332             if(!info.isShardReadyWithLeaderId()){
 333                 isReady = false;
 334                 break;
 335             }
 336         }
 337         return isReady;
 338     }
 339
 340     private boolean isInSync(){
 341         for (ShardInformation info : localShards.values()) {
 342             if(!info.isInSync()){
 343                 return false;
 344             }
 345         }
 346         return true;
 347     }
 348
 349     private void onActorInitialized(Object message) {
 350         final ActorRef sender = getSender();
 351
 352         if (sender == null) {
 353             return; //why is a non-actor sending this message? Just ignore.
 354         }
 355
 356         String actorName = sender.path().name();
 357         //find shard name from actor name; actor name is stringified shardId
 358         ShardIdentifier shardId = ShardIdentifier.builder().fromShardIdString(actorName).build();
 359
 360         if (shardId.getShardName() == null) {
 361             return;
 362         }
 363
 364         markShardAsInitialized(shardId.getShardName());
 365     }
 366
 367     private void markShardAsInitialized(String shardName) {
 368         LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
 369
 370         ShardInformation shardInformation = localShards.get(shardName);
 371         if (shardInformation != null) {
 372             shardInformation.setActorInitialized();
 373
 374             shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
 375         }
 376     }
 377
 378     @Override
 379     protected void handleRecover(Object message) throws Exception {
 380         if (message instanceof RecoveryCompleted) {
 381             LOG.info("Recovery complete : {}", persistenceId());
 382
 383             // We no longer persist SchemaContext modules so delete all the prior messages from the akka
 384             // journal on upgrade from Helium.
 385             deleteMessages(lastSequenceNr());
 386         }
 387     }
 388
 389     private void findLocalShard(FindLocalShard message) {
 390         final ShardInformation shardInformation = localShards.get(message.getShardName());
 391
 392         if(shardInformation == null){
 393             getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
 394             return;
 395         }
 396
 397         sendResponse(shardInformation, message.isWaitUntilInitialized(), false, new Supplier<Object>() {
 398             @Override
 399             public Object get() {
 400                 return new LocalShardFound(shardInformation.getActor());
 401             }
 402         });
 403     }
 404
 405     private void sendResponse(ShardInformation shardInformation, boolean doWait,
 406             boolean wantShardReady, final Supplier<Object> messageSupplier) {
 407         if (!shardInformation.isShardInitialized() || (wantShardReady && !shardInformation.isShardReadyWithLeaderId())) {
 408             if(doWait) {
 409                 final ActorRef sender = getSender();
 410                 final ActorRef self = self();
 411
 412                 Runnable replyRunnable = new Runnable() {
 413                     @Override
 414                     public void run() {
 415                         sender.tell(messageSupplier.get(), self);
 416                     }
 417                 };
 418
 419                 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
 420                     new OnShardInitialized(replyRunnable);
 421
 422                 shardInformation.addOnShardInitialized(onShardInitialized);
 423
 424                 LOG.debug("{}: Scheduling timer to wait for shard {}", persistenceId(), shardInformation.getShardName());
 425
 426                 FiniteDuration timeout = datastoreContext.getShardInitializationTimeout().duration();
 427                 if(shardInformation.isShardInitialized()) {
 428                     // If the shard is already initialized then we'll wait enough time for the shard to
 429                     // elect a leader, ie 2 times the election timeout.
 430                     timeout = FiniteDuration.create(datastoreContext.getShardRaftConfig()
 431                             .getElectionTimeOutInterval().toMillis() * 2, TimeUnit.MILLISECONDS);
 432                 }
 433
 434                 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
 435                         timeout, getSelf(),
 436                         new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
 437                         getContext().dispatcher(), getSelf());
 438
 439                 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
 440
 441             } else if (!shardInformation.isShardInitialized()) {
 442                 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
 443                         shardInformation.getShardName());
 444                 getSender().tell(createNotInitializedException(shardInformation.shardId), getSelf());
 445             } else {
 446                 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
 447                         shardInformation.getShardName());
 448                 getSender().tell(createNoShardLeaderException(shardInformation.shardId), getSelf());
 449             }
 450
 451             return;
 452         }
 453
 454         getSender().tell(messageSupplier.get(), getSelf());
 455     }
 456
 457     private NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) {
 458         return new NoShardLeaderException(null, shardId.toString());
 459     }
 460
 461     private NotInitializedException createNotInitializedException(ShardIdentifier shardId) {
 462         return new NotInitializedException(String.format(
 463                 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
 464     }
 465
 466     private void memberRemoved(ClusterEvent.MemberRemoved message) {
 467         String memberName = message.member().roles().head();
 468
 469         LOG.debug("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
 470                 message.member().address());
 471
 472         memberNameToAddress.remove(memberName);
 473
 474         for(ShardInformation info : localShards.values()){
 475             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 476         }
 477     }
 478
 479     private void memberExited(ClusterEvent.MemberExited message) {
 480         String memberName = message.member().roles().head();
 481
 482         LOG.debug("{}: Received MemberExited: memberName: {}, address: {}", persistenceId(), memberName,
 483                 message.member().address());
 484
 485         memberNameToAddress.remove(memberName);
 486
 487         for(ShardInformation info : localShards.values()){
 488             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 489         }
 490     }
 491
 492     private void memberUp(ClusterEvent.MemberUp message) {
 493         String memberName = message.member().roles().head();
 494
 495         LOG.debug("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
 496                 message.member().address());
 497
 498         memberNameToAddress.put(memberName, message.member().address());
 499
 500         for(ShardInformation info : localShards.values()){
 501             String shardName = info.getShardName();
 502             String peerId = getShardIdentifier(memberName, shardName).toString();
 503             info.updatePeerAddress(peerId, getShardActorPath(shardName, memberName), getSelf());
 504
 505             info.peerUp(memberName, peerId, getSelf());
 506         }
 507
 508         checkReady();
 509     }
 510
 511     private void memberReachable(ClusterEvent.ReachableMember message) {
 512         String memberName = message.member().roles().head();
 513         LOG.debug("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
 514
 515         markMemberAvailable(memberName);
 516     }
 517
 518     private void memberUnreachable(ClusterEvent.UnreachableMember message) {
 519         String memberName = message.member().roles().head();
 520         LOG.debug("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
 521
 522         markMemberUnavailable(memberName);
 523     }
 524
 525     private void markMemberUnavailable(final String memberName) {
 526         for(ShardInformation info : localShards.values()){
 527             String leaderId = info.getLeaderId();
 528             if(leaderId != null && leaderId.contains(memberName)) {
 529                 LOG.debug("Marking Leader {} as unavailable.", leaderId);
 530                 info.setLeaderAvailable(false);
 531
 532                 primaryShardInfoCache.remove(info.getShardName());
 533             }
 534
 535             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 536         }
 537     }
 538
 539     private void markMemberAvailable(final String memberName) {
 540         for(ShardInformation info : localShards.values()){
 541             String leaderId = info.getLeaderId();
 542             if(leaderId != null && leaderId.contains(memberName)) {
 543                 LOG.debug("Marking Leader {} as available.", leaderId);
 544                 info.setLeaderAvailable(true);
 545             }
 546
 547             info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 548         }
 549     }
 550
 551     private void onDatastoreContext(DatastoreContext context) {
 552         datastoreContext = context;
 553         for (ShardInformation info : localShards.values()) {
 554             if (info.getActor() != null) {
 555                 info.getActor().tell(datastoreContext, getSelf());
 556             }
 557         }
 558     }
 559
 560     private void onSwitchShardBehavior(SwitchShardBehavior message) {
 561         ShardIdentifier identifier = ShardIdentifier.builder().fromShardIdString(message.getShardName()).build();
 562
 563         ShardInformation shardInformation = localShards.get(identifier.getShardName());
 564
 565         if(shardInformation != null && shardInformation.getActor() != null) {
 566             shardInformation.getActor().tell(
 567                     new SwitchBehavior(RaftState.valueOf(message.getNewState()), message.getTerm()), getSelf());
 568         } else {
 569             LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
 570                     message.getShardName(), message.getNewState());
 571         }
 572     }
 573
 574     /**
 575      * Notifies all the local shards of a change in the schema context
 576      *
 577      * @param message
 578      */
 579     private void updateSchemaContext(final Object message) {
 580         schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
 581
 582         LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getAllModuleIdentifiers().size());
 583
 584         for (ShardInformation info : localShards.values()) {
 585             if (info.getActor() == null) {
 586                 LOG.debug("Creating Shard {}", info.getShardId());
 587                 info.setActor(newShardActor(schemaContext, info));
 588             } else {
 589                 info.getActor().tell(message, getSelf());
 590             }
 591         }
 592     }
 593
 594     @VisibleForTesting
 595     protected ClusterWrapper getCluster() {
 596         return cluster;
 597     }
 598
 599     @VisibleForTesting
 600     protected ActorRef newShardActor(final SchemaContext schemaContext, ShardInformation info) {
 601         return getContext().actorOf(info.newProps(schemaContext)
 602                         .withDispatcher(shardDispatcherPath), info.getShardId().toString());
 603     }
 604
 605     private void findPrimary(FindPrimary message) {
 606         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
 607
 608         final String shardName = message.getShardName();
 609         final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
 610
 611         // First see if the there is a local replica for the shard
 612         final ShardInformation info = localShards.get(shardName);
 613         if (info != null) {
 614             sendResponse(info, message.isWaitUntilReady(), true, new Supplier<Object>() {
 615                 @Override
 616                 public Object get() {
 617                     String primaryPath = info.getSerializedLeaderActor();
 618                     Object found = canReturnLocalShardState && info.isLeader() ?
 619                             new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
 620                                 new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
 621
 622                     if(LOG.isDebugEnabled()) {
 623                         LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
 624                     }
 625
 626                     return found;
 627                 }
 628             });
 629
 630             return;
 631         }
 632
 633         for(Map.Entry<String, Address> entry: memberNameToAddress.entrySet()) {
 634             if(!cluster.getCurrentMemberName().equals(entry.getKey())) {
 635                 String path = getShardManagerActorPathBuilder(entry.getValue()).toString();
 636
 637                 LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}", persistenceId(),
 638                         shardName, path);
 639
 640                 getContext().actorSelection(path).forward(new RemoteFindPrimary(shardName,
 641                         message.isWaitUntilReady()), getContext());
 642                 return;
 643             }
 644         }
 645
 646         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
 647
 648         getSender().tell(new PrimaryNotFoundException(
 649                 String.format("No primary shard found for %s.", shardName)), getSelf());
 650     }
 651
 652     private StringBuilder getShardManagerActorPathBuilder(Address address) {
 653         StringBuilder builder = new StringBuilder();
 654         builder.append(address.toString()).append("/user/").append(shardManagerIdentifierString);
 655         return builder;
 656     }
 657
 658     private String getShardActorPath(String shardName, String memberName) {
 659         Address address = memberNameToAddress.get(memberName);
 660         if(address != null) {
 661             StringBuilder builder = getShardManagerActorPathBuilder(address);
 662             builder.append("/")
 663                 .append(getShardIdentifier(memberName, shardName));
 664             return builder.toString();
 665         }
 666         return null;
 667     }
 668
 669     /**
 670      * Construct the name of the shard actor given the name of the member on
 671      * which the shard resides and the name of the shard
 672      *
 673      * @param memberName
 674      * @param shardName
 675      * @return
 676      */
 677     private ShardIdentifier getShardIdentifier(String memberName, String shardName){
 678         return ShardIdentifier.builder().memberName(memberName).shardName(shardName).type(type).build();
 679     }
 680
 681     /**
 682      * Create shards that are local to the member on which the ShardManager
 683      * runs
 684      *
 685      */
 686     private void createLocalShards() {
 687         String memberName = this.cluster.getCurrentMemberName();
 688         Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
 689
 690         ShardPropsCreator shardPropsCreator = new DefaultShardPropsCreator();
 691         List<String> localShardActorNames = new ArrayList<>();
 692         for(String shardName : memberShardNames){
 693             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 694             Map<String, String> peerAddresses = getPeerAddresses(shardName);
 695             localShardActorNames.add(shardId.toString());
 696             localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses, datastoreContext,
 697                     shardPropsCreator));
 698         }
 699
 700         mBean = ShardManagerInfo.createShardManagerMBean(memberName, "shard-manager-" + this.type,
 701                     datastoreContext.getDataStoreMXBeanType(), localShardActorNames);
 702
 703         mBean.setShardManager(this);
 704     }
 705
 706     /**
 707      * Given the name of the shard find the addresses of all it's peers
 708      *
 709      * @param shardName
 710      * @return
 711      */
 712     private Map<String, String> getPeerAddresses(String shardName) {
 713         return getPeerAddresses(shardName, configuration.getMembersFromShardName(shardName));
 714     }
 715
 716     private Map<String, String> getPeerAddresses(String shardName, Collection<String> members) {
 717
 718         Map<String, String> peerAddresses = new HashMap<>();
 719
 720         String currentMemberName = this.cluster.getCurrentMemberName();
 721
 722         for(String memberName : members) {
 723             if(!currentMemberName.equals(memberName)) {
 724                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 725                 String path = getShardActorPath(shardName, memberName);
 726                 peerAddresses.put(shardId.toString(), path);
 727             }
 728         }
 729         return peerAddresses;
 730     }
 731
 732     @Override
 733     public SupervisorStrategy supervisorStrategy() {
 734
 735         return new OneForOneStrategy(10, Duration.create("1 minute"),
 736             new Function<Throwable, SupervisorStrategy.Directive>() {
 737                 @Override
 738                 public SupervisorStrategy.Directive apply(Throwable t) {
 739                     LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
 740                     return SupervisorStrategy.resume();
 741                 }
 742             }
 743         );
 744
 745     }
 746
 747     @Override
 748     public String persistenceId() {
 749         return "shard-manager-" + type;
 750     }
 751
 752     @VisibleForTesting
 753     ShardManagerInfoMBean getMBean(){
 754         return mBean;
 755     }
 756
 757     @VisibleForTesting
 758     protected static class ShardInformation {
 759         private final ShardIdentifier shardId;
 760         private final String shardName;
 761         private ActorRef actor;
 762         private ActorPath actorPath;
 763         private final Map<String, String> peerAddresses;
 764         private Optional<DataTree> localShardDataTree;
 765         private boolean leaderAvailable = false;
 766
 767         // flag that determines if the actor is ready for business
 768         private boolean actorInitialized = false;
 769
 770         private boolean followerSyncStatus = false;
 771
 772         private final Set<OnShardInitialized> onShardInitializedSet = Sets.newHashSet();
 773         private String role ;
 774         private String leaderId;
 775         private short leaderVersion;
 776
 777         private final DatastoreContext datastoreContext;
 778         private final ShardPropsCreator shardPropsCreator;
 779
 780         private ShardInformation(String shardName, ShardIdentifier shardId,
 781                 Map<String, String> peerAddresses, DatastoreContext datastoreContext,
 782                 ShardPropsCreator shardPropsCreator) {
 783             this.shardName = shardName;
 784             this.shardId = shardId;
 785             this.peerAddresses = peerAddresses;
 786             this.datastoreContext = datastoreContext;
 787             this.shardPropsCreator = shardPropsCreator;
 788         }
 789
 790         Props newProps(SchemaContext schemaContext) {
 791             return shardPropsCreator.newProps(shardId, peerAddresses, datastoreContext, schemaContext);
 792         }
 793
 794         String getShardName() {
 795             return shardName;
 796         }
 797
 798         ActorRef getActor(){
 799             return actor;
 800         }
 801
 802         ActorPath getActorPath() {
 803             return actorPath;
 804         }
 805
 806         void setActor(ActorRef actor) {
 807             this.actor = actor;
 808             this.actorPath = actor.path();
 809         }
 810
 811         ShardIdentifier getShardId() {
 812             return shardId;
 813         }
 814
 815         void setLocalDataTree(Optional<DataTree> localShardDataTree) {
 816             this.localShardDataTree = localShardDataTree;
 817         }
 818
 819         Optional<DataTree> getLocalShardDataTree() {
 820             return localShardDataTree;
 821         }
 822
 823         Map<String, String> getPeerAddresses() {
 824             return peerAddresses;
 825         }
 826
 827         void updatePeerAddress(String peerId, String peerAddress, ActorRef sender){
 828             LOG.info("updatePeerAddress for peer {} with address {}", peerId,
 829                 peerAddress);
 830             if(peerAddresses.containsKey(peerId)){
 831                 peerAddresses.put(peerId, peerAddress);
 832
 833                 if(actor != null) {
 834                     if(LOG.isDebugEnabled()) {
 835                         LOG.debug("Sending PeerAddressResolved for peer {} with address {} to {}",
 836                                 peerId, peerAddress, actor.path());
 837                     }
 838
 839                     actor.tell(new PeerAddressResolved(peerId, peerAddress), sender);
 840                 }
 841
 842                 notifyOnShardInitializedCallbacks();
 843             }
 844         }
 845
 846         void peerDown(String memberName, String peerId, ActorRef sender) {
 847             if(peerAddresses.containsKey(peerId) && actor != null) {
 848                 actor.tell(new PeerDown(memberName, peerId), sender);
 849             }
 850         }
 851
 852         void peerUp(String memberName, String peerId, ActorRef sender) {
 853             if(peerAddresses.containsKey(peerId) && actor != null) {
 854                 actor.tell(new PeerUp(memberName, peerId), sender);
 855             }
 856         }
 857
 858         boolean isShardReady() {
 859             return !RaftState.Candidate.name().equals(role) && !Strings.isNullOrEmpty(role);
 860         }
 861
 862         boolean isShardReadyWithLeaderId() {
 863             return leaderAvailable && isShardReady() && !RaftState.IsolatedLeader.name().equals(role) &&
 864                     (isLeader() || peerAddresses.get(leaderId) != null);
 865         }
 866
 867         boolean isShardInitialized() {
 868             return getActor() != null && actorInitialized;
 869         }
 870
 871         boolean isLeader() {
 872             return Objects.equal(leaderId, shardId.toString());
 873         }
 874
 875         String getSerializedLeaderActor() {
 876             if(isLeader()) {
 877                 return Serialization.serializedActorPath(getActor());
 878             } else {
 879                 return peerAddresses.get(leaderId);
 880             }
 881         }
 882
 883         void setActorInitialized() {
 884             LOG.debug("Shard {} is initialized", shardId);
 885
 886             this.actorInitialized = true;
 887
 888             notifyOnShardInitializedCallbacks();
 889         }
 890
 891         private void notifyOnShardInitializedCallbacks() {
 892             if(onShardInitializedSet.isEmpty()) {
 893                 return;
 894             }
 895
 896             boolean ready = isShardReadyWithLeaderId();
 897
 898             if(LOG.isDebugEnabled()) {
 899                 LOG.debug("Shard {} is {} - notifying {} OnShardInitialized callbacks", shardId,
 900                         ready ? "ready" : "initialized", onShardInitializedSet.size());
 901             }
 902
 903             Iterator<OnShardInitialized> iter = onShardInitializedSet.iterator();
 904             while(iter.hasNext()) {
 905                 OnShardInitialized onShardInitialized = iter.next();
 906                 if(!(onShardInitialized instanceof OnShardReady) || ready) {
 907                     iter.remove();
 908                     onShardInitialized.getTimeoutSchedule().cancel();
 909                     onShardInitialized.getReplyRunnable().run();
 910                 }
 911             }
 912         }
 913
 914         void addOnShardInitialized(OnShardInitialized onShardInitialized) {
 915             onShardInitializedSet.add(onShardInitialized);
 916         }
 917
 918         void removeOnShardInitialized(OnShardInitialized onShardInitialized) {
 919             onShardInitializedSet.remove(onShardInitialized);
 920         }
 921
 922         void setRole(String newRole) {
 923             this.role = newRole;
 924
 925             notifyOnShardInitializedCallbacks();
 926         }
 927
 928         void setFollowerSyncStatus(boolean syncStatus){
 929             this.followerSyncStatus = syncStatus;
 930         }
 931
 932         boolean isInSync(){
 933             if(RaftState.Follower.name().equals(this.role)){
 934                 return followerSyncStatus;
 935             } else if(RaftState.Leader.name().equals(this.role)){
 936                 return true;
 937             }
 938
 939             return false;
 940         }
 941
 942         boolean setLeaderId(String leaderId) {
 943             boolean changed = !Objects.equal(this.leaderId, leaderId);
 944             this.leaderId = leaderId;
 945             if(leaderId != null) {
 946                 this.leaderAvailable = true;
 947             }
 948             notifyOnShardInitializedCallbacks();
 949
 950             return changed;
 951         }
 952
 953         String getLeaderId() {
 954             return leaderId;
 955         }
 956
 957         void setLeaderAvailable(boolean leaderAvailable) {
 958             this.leaderAvailable = leaderAvailable;
 959         }
 960
 961         short getLeaderVersion() {
 962             return leaderVersion;
 963         }
 964
 965         void setLeaderVersion(short leaderVersion) {
 966             this.leaderVersion = leaderVersion;
 967         }
 968     }
 969
 970     private static class ShardManagerCreator implements Creator<ShardManager> {
 971         private static final long serialVersionUID = 1L;
 972
 973         final ClusterWrapper cluster;
 974         final Configuration configuration;
 975         final DatastoreContext datastoreContext;
 976         private final CountDownLatch waitTillReadyCountdownLatch;
 977         private final PrimaryShardInfoFutureCache primaryShardInfoCache;
 978
 979         ShardManagerCreator(ClusterWrapper cluster, Configuration configuration, DatastoreContext datastoreContext,
 980                 CountDownLatch waitTillReadyCountdownLatch, PrimaryShardInfoFutureCache primaryShardInfoCache) {
 981             this.cluster = cluster;
 982             this.configuration = configuration;
 983             this.datastoreContext = datastoreContext;
 984             this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
 985             this.primaryShardInfoCache = primaryShardInfoCache;
 986         }
 987
 988         @Override
 989         public ShardManager create() throws Exception {
 990             return new ShardManager(cluster, configuration, datastoreContext, waitTillReadyCountdownLatch,
 991                     primaryShardInfoCache);
 992         }
 993     }
 994
 995     private static class OnShardInitialized {
 996         private final Runnable replyRunnable;
 997         private Cancellable timeoutSchedule;
 998
 999         OnShardInitialized(Runnable replyRunnable) {
1000             this.replyRunnable = replyRunnable;
1001         }
1002
1003         Runnable getReplyRunnable() {
1004             return replyRunnable;
1005         }
1006
1007         Cancellable getTimeoutSchedule() {
1008             return timeoutSchedule;
1009         }
1010
1011         void setTimeoutSchedule(Cancellable timeoutSchedule) {
1012             this.timeoutSchedule = timeoutSchedule;
1013         }
1014     }
1015
1016     private static class OnShardReady extends OnShardInitialized {
1017         OnShardReady(Runnable replyRunnable) {
1018             super(replyRunnable);
1019         }
1020     }
1021
1022     private static class ShardNotInitializedTimeout {
1023         private final ActorRef sender;
1024         private final ShardInformation shardInfo;
1025         private final OnShardInitialized onShardInitialized;
1026
1027         ShardNotInitializedTimeout(ShardInformation shardInfo, OnShardInitialized onShardInitialized, ActorRef sender) {
1028             this.sender = sender;
1029             this.shardInfo = shardInfo;
1030             this.onShardInitialized = onShardInitialized;
1031         }
1032
1033         ActorRef getSender() {
1034             return sender;
1035         }
1036
1037         ShardInformation getShardInfo() {
1038             return shardInfo;
1039         }
1040
1041         OnShardInitialized getOnShardInitialized() {
1042             return onShardInitialized;
1043         }
1044     }
1045
1046     /**
1047      * We no longer persist SchemaContextModules but keep this class around for now for backwards
1048      * compatibility so we don't get de-serialization failures on upgrade from Helium.
1049      */
1050     @Deprecated
1051     static class SchemaContextModules implements Serializable {
1052         private static final long serialVersionUID = -8884620101025936590L;
1053
1054         private final Set<String> modules;
1055
1056         SchemaContextModules(Set<String> modules){
1057             this.modules = modules;
1058         }
1059
1060         public Set<String> getModules() {
1061             return modules;
1062         }
1063     }
1064 }
1065
1066
1067