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