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