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 onLeaderStateChanged(LeaderStateChanged leaderStateChanged) {
 197         LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
 198
 199         ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
 200         if(shardInformation != null) {
 201             shardInformation.setLeaderId(leaderStateChanged.getLeaderId());
 202             if (isReadyWithLeaderId()) {
 203                 LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
 204                         persistenceId(), type, waitTillReadyCountdownLatch.getCount());
 205
 206                 waitTillReadyCountdownLatch.countDown();
 207             }
 208
 209         } else {
 210             LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
 211         }
 212     }
 213
 214     private void onShardNotInitializedTimeout(ShardNotInitializedTimeout message) {
 215         ShardInformation shardInfo = message.getShardInfo();
 216
 217         LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
 218                 shardInfo.getShardName());
 219
 220         shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
 221
 222         if(!shardInfo.isShardInitialized()) {
 223             LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
 224             message.getSender().tell(createNotInitializedException(shardInfo.shardId), getSelf());
 225         } else {
 226             LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
 227             message.getSender().tell(createNoShardLeaderException(shardInfo.shardId), getSelf());
 228         }
 229     }
 230
 231     private void onFollowerInitialSyncStatus(FollowerInitialSyncUpStatus status) {
 232         LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
 233                 status.getName(), status.isInitialSyncDone());
 234
 235         ShardInformation shardInformation = findShardInformation(status.getName());
 236
 237         if(shardInformation != null) {
 238             shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
 239
 240             mBean.setSyncStatus(isInSync());
 241         }
 242
 243     }
 244
 245     private void onRoleChangeNotification(RoleChangeNotification roleChanged) {
 246         LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
 247                 roleChanged.getOldRole(), roleChanged.getNewRole());
 248
 249         ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
 250         if(shardInformation != null) {
 251             shardInformation.setRole(roleChanged.getNewRole());
 252
 253             if (isReadyWithLeaderId()) {
 254                 LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
 255                         persistenceId(), type, waitTillReadyCountdownLatch.getCount());
 256
 257                 waitTillReadyCountdownLatch.countDown();
 258             }
 259
 260             mBean.setSyncStatus(isInSync());
 261         }
 262     }
 263
 264
 265     private ShardInformation findShardInformation(String memberId) {
 266         for(ShardInformation info : localShards.values()){
 267             if(info.getShardId().toString().equals(memberId)){
 268                 return info;
 269             }
 270         }
 271
 272         return null;
 273     }
 274
 275     private boolean isReadyWithLeaderId() {
 276         boolean isReady = true;
 277         for (ShardInformation info : localShards.values()) {
 278             if(!info.isShardReadyWithLeaderId()){
 279                 isReady = false;
 280                 break;
 281             }
 282         }
 283         return isReady;
 284     }
 285
 286     private boolean isInSync(){
 287         for (ShardInformation info : localShards.values()) {
 288             if(!info.isInSync()){
 289                 return false;
 290             }
 291         }
 292         return true;
 293     }
 294
 295     private void onActorInitialized(Object message) {
 296         final ActorRef sender = getSender();
 297
 298         if (sender == null) {
 299             return; //why is a non-actor sending this message? Just ignore.
 300         }
 301
 302         String actorName = sender.path().name();
 303         //find shard name from actor name; actor name is stringified shardId
 304         ShardIdentifier shardId = ShardIdentifier.builder().fromShardIdString(actorName).build();
 305
 306         if (shardId.getShardName() == null) {
 307             return;
 308         }
 309
 310         markShardAsInitialized(shardId.getShardName());
 311     }
 312
 313     private void markShardAsInitialized(String shardName) {
 314         LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
 315
 316         ShardInformation shardInformation = localShards.get(shardName);
 317         if (shardInformation != null) {
 318             shardInformation.setActorInitialized();
 319
 320             shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
 321         }
 322     }
 323
 324     @Override
 325     protected void handleRecover(Object message) throws Exception {
 326         if(dataPersistenceProvider.isRecoveryApplicable()) {
 327             if (message instanceof SchemaContextModules) {
 328                 SchemaContextModules msg = (SchemaContextModules) message;
 329                 knownModules = ImmutableSet.copyOf(msg.getModules());
 330             } else if (message instanceof RecoveryFailure) {
 331                 RecoveryFailure failure = (RecoveryFailure) message;
 332                 LOG.error("Recovery failed", failure.cause());
 333             } else if (message instanceof RecoveryCompleted) {
 334                 LOG.info("Recovery complete : {}", persistenceId());
 335
 336                 // Delete all the messages from the akka journal except the last one
 337                 deleteMessages(lastSequenceNr() - 1);
 338             }
 339         } else {
 340             if (message instanceof RecoveryCompleted) {
 341                 LOG.info("Recovery complete : {}", persistenceId());
 342
 343                 // Delete all the messages from the akka journal
 344                 deleteMessages(lastSequenceNr());
 345             }
 346         }
 347     }
 348
 349     private void findLocalShard(FindLocalShard message) {
 350         final ShardInformation shardInformation = localShards.get(message.getShardName());
 351
 352         if(shardInformation == null){
 353             getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
 354             return;
 355         }
 356
 357         sendResponse(shardInformation, message.isWaitUntilInitialized(), false, new Supplier<Object>() {
 358             @Override
 359             public Object get() {
 360                 return new LocalShardFound(shardInformation.getActor());
 361             }
 362         });
 363     }
 364
 365     private void sendResponse(ShardInformation shardInformation, boolean doWait,
 366             boolean wantShardReady, final Supplier<Object> messageSupplier) {
 367         if (!shardInformation.isShardInitialized() || (wantShardReady && !shardInformation.isShardReadyWithLeaderId())) {
 368             if(doWait) {
 369                 final ActorRef sender = getSender();
 370                 final ActorRef self = self();
 371
 372                 Runnable replyRunnable = new Runnable() {
 373                     @Override
 374                     public void run() {
 375                         sender.tell(messageSupplier.get(), self);
 376                     }
 377                 };
 378
 379                 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
 380                     new OnShardInitialized(replyRunnable);
 381
 382                 shardInformation.addOnShardInitialized(onShardInitialized);
 383
 384                 LOG.debug("{}: Scheduling timer to wait for shard {}", persistenceId(), shardInformation.getShardName());
 385
 386                 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
 387                         datastoreContext.getShardInitializationTimeout().duration(), getSelf(),
 388                         new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
 389                         getContext().dispatcher(), getSelf());
 390
 391                 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
 392
 393             } else if (!shardInformation.isShardInitialized()) {
 394                 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
 395                         shardInformation.getShardName());
 396                 getSender().tell(createNotInitializedException(shardInformation.shardId), getSelf());
 397             } else {
 398                 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
 399                         shardInformation.getShardName());
 400                 getSender().tell(createNoShardLeaderException(shardInformation.shardId), getSelf());
 401             }
 402
 403             return;
 404         }
 405
 406         getSender().tell(messageSupplier.get(), getSelf());
 407     }
 408
 409     private NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) {
 410         return new NoShardLeaderException(String.format(
 411                 "Could not find a leader for shard %s. This typically happens when the system is coming up or " +
 412                 "recovering and a leader is being elected. Try again later.", shardId));
 413     }
 414
 415     private NotInitializedException createNotInitializedException(ShardIdentifier shardId) {
 416         return new NotInitializedException(String.format(
 417                 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
 418     }
 419
 420     private void memberRemoved(ClusterEvent.MemberRemoved message) {
 421         String memberName = message.member().roles().head();
 422
 423         LOG.debug("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
 424                 message.member().address());
 425
 426         memberNameToAddress.remove(message.member().roles().head());
 427     }
 428
 429     private void memberUp(ClusterEvent.MemberUp message) {
 430         String memberName = message.member().roles().head();
 431
 432         LOG.debug("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
 433                 message.member().address());
 434
 435         memberNameToAddress.put(memberName, message.member().address());
 436
 437         for(ShardInformation info : localShards.values()){
 438             String shardName = info.getShardName();
 439             info.updatePeerAddress(getShardIdentifier(memberName, shardName).toString(),
 440                 getShardActorPath(shardName, memberName), getSelf());
 441         }
 442     }
 443
 444     private void onDatastoreContext(DatastoreContext context) {
 445         datastoreContext = context;
 446         for (ShardInformation info : localShards.values()) {
 447             if (info.getActor() != null) {
 448                 info.getActor().tell(datastoreContext, getSelf());
 449             }
 450         }
 451     }
 452
 453     /**
 454      * Notifies all the local shards of a change in the schema context
 455      *
 456      * @param message
 457      */
 458     private void updateSchemaContext(final Object message) {
 459         final SchemaContext schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
 460
 461         Set<ModuleIdentifier> allModuleIdentifiers = schemaContext.getAllModuleIdentifiers();
 462         Set<String> newModules = new HashSet<>(128);
 463
 464         for(ModuleIdentifier moduleIdentifier : allModuleIdentifiers){
 465             String s = moduleIdentifier.getNamespace().toString();
 466             newModules.add(s);
 467         }
 468
 469         if(newModules.containsAll(knownModules)) {
 470
 471             LOG.debug("New SchemaContext has a super set of current knownModules - persisting info");
 472
 473             knownModules = ImmutableSet.copyOf(newModules);
 474
 475             dataPersistenceProvider.persist(new SchemaContextModules(newModules), new Procedure<SchemaContextModules>() {
 476
 477                 @Override
 478                 public void apply(SchemaContextModules param) throws Exception {
 479                     LOG.debug("Sending new SchemaContext to Shards");
 480                     for (ShardInformation info : localShards.values()) {
 481                         if (info.getActor() == null) {
 482                             info.setActor(newShardActor(schemaContext, info));
 483                         } else {
 484                             info.getActor().tell(message, getSelf());
 485                         }
 486                     }
 487                 }
 488
 489             });
 490         } else {
 491             LOG.debug("Rejecting schema context update - not a super set of previously known modules:\nUPDATE: {}\nKNOWN: {}",
 492                     newModules, knownModules);
 493         }
 494
 495     }
 496
 497     @VisibleForTesting
 498     protected ClusterWrapper getCluster() {
 499         return cluster;
 500     }
 501
 502     @VisibleForTesting
 503     protected ActorRef newShardActor(final SchemaContext schemaContext, ShardInformation info) {
 504         return getContext().actorOf(Shard.props(info.getShardId(),
 505                 info.getPeerAddresses(), datastoreContext, schemaContext)
 506                         .withDispatcher(shardDispatcherPath), info.getShardId().toString());
 507     }
 508
 509     private void findPrimary(FindPrimary message) {
 510         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
 511
 512         final String shardName = message.getShardName();
 513
 514         // First see if the there is a local replica for the shard
 515         final ShardInformation info = localShards.get(shardName);
 516         if (info != null) {
 517             sendResponse(info, message.isWaitUntilReady(), true, new Supplier<Object>() {
 518                 @Override
 519                 public Object get() {
 520                     Object found = new PrimaryFound(info.getSerializedLeaderActor());
 521
 522                     if(LOG.isDebugEnabled()) {
 523                         LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
 524                     }
 525
 526                     return found;
 527                 }
 528             });
 529
 530             return;
 531         }
 532
 533         for(Map.Entry<String, Address> entry: memberNameToAddress.entrySet()) {
 534             if(!cluster.getCurrentMemberName().equals(entry.getKey())) {
 535                 String path = getShardManagerActorPathBuilder(entry.getValue()).toString();
 536
 537                 LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}", persistenceId(),
 538                         shardName, path);
 539
 540                 getContext().actorSelection(path).forward(message, getContext());
 541                 return;
 542             }
 543         }
 544
 545         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
 546
 547         getSender().tell(new PrimaryNotFoundException(
 548                 String.format("No primary shard found for %s.", shardName)), getSelf());
 549     }
 550
 551     private StringBuilder getShardManagerActorPathBuilder(Address address) {
 552         StringBuilder builder = new StringBuilder();
 553         builder.append(address.toString()).append("/user/").append(shardManagerIdentifierString);
 554         return builder;
 555     }
 556
 557     private String getShardActorPath(String shardName, String memberName) {
 558         Address address = memberNameToAddress.get(memberName);
 559         if(address != null) {
 560             StringBuilder builder = getShardManagerActorPathBuilder(address);
 561             builder.append("/")
 562                 .append(getShardIdentifier(memberName, shardName));
 563             return builder.toString();
 564         }
 565         return null;
 566     }
 567
 568     /**
 569      * Construct the name of the shard actor given the name of the member on
 570      * which the shard resides and the name of the shard
 571      *
 572      * @param memberName
 573      * @param shardName
 574      * @return
 575      */
 576     private ShardIdentifier getShardIdentifier(String memberName, String shardName){
 577         return ShardIdentifier.builder().memberName(memberName).shardName(shardName).type(type).build();
 578     }
 579
 580     /**
 581      * Create shards that are local to the member on which the ShardManager
 582      * runs
 583      *
 584      */
 585     private void createLocalShards() {
 586         String memberName = this.cluster.getCurrentMemberName();
 587         List<String> memberShardNames =
 588             this.configuration.getMemberShardNames(memberName);
 589
 590         List<String> localShardActorNames = new ArrayList<>();
 591         for(String shardName : memberShardNames){
 592             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 593             Map<String, String> peerAddresses = getPeerAddresses(shardName);
 594             localShardActorNames.add(shardId.toString());
 595             localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses));
 596         }
 597
 598         mBean = ShardManagerInfo.createShardManagerMBean("shard-manager-" + this.type,
 599                     datastoreContext.getDataStoreMXBeanType(), localShardActorNames);
 600     }
 601
 602     /**
 603      * Given the name of the shard find the addresses of all it's peers
 604      *
 605      * @param shardName
 606      * @return
 607      */
 608     private Map<String, String> getPeerAddresses(String shardName){
 609
 610         Map<String, String> peerAddresses = new HashMap<>();
 611
 612         List<String> members = this.configuration.getMembersFromShardName(shardName);
 613
 614         String currentMemberName = this.cluster.getCurrentMemberName();
 615
 616         for(String memberName : members){
 617             if(!currentMemberName.equals(memberName)){
 618                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 619                 String path = getShardActorPath(shardName, currentMemberName);
 620                 peerAddresses.put(shardId.toString(), path);
 621             }
 622         }
 623         return peerAddresses;
 624     }
 625
 626     @Override
 627     public SupervisorStrategy supervisorStrategy() {
 628
 629         return new OneForOneStrategy(10, Duration.create("1 minute"),
 630             new Function<Throwable, SupervisorStrategy.Directive>() {
 631                 @Override
 632                 public SupervisorStrategy.Directive apply(Throwable t) {
 633                     LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
 634                     return SupervisorStrategy.resume();
 635                 }
 636             }
 637         );
 638
 639     }
 640
 641     @Override
 642     public String persistenceId() {
 643         return "shard-manager-" + type;
 644     }
 645
 646     @VisibleForTesting
 647     Collection<String> getKnownModules() {
 648         return knownModules;
 649     }
 650
 651     @VisibleForTesting
 652     DataPersistenceProvider getDataPersistenceProvider() {
 653         return dataPersistenceProvider;
 654     }
 655
 656     @VisibleForTesting
 657     ShardManagerInfoMBean getMBean(){
 658         return mBean;
 659     }
 660
 661     @VisibleForTesting
 662     protected static class ShardInformation {
 663         private final ShardIdentifier shardId;
 664         private final String shardName;
 665         private ActorRef actor;
 666         private ActorPath actorPath;
 667         private final Map<String, String> peerAddresses;
 668
 669         // flag that determines if the actor is ready for business
 670         private boolean actorInitialized = false;
 671
 672         private boolean followerSyncStatus = false;
 673
 674         private final Set<OnShardInitialized> onShardInitializedSet = Sets.newHashSet();
 675         private String role ;
 676         private String leaderId;
 677
 678         private ShardInformation(String shardName, ShardIdentifier shardId,
 679                 Map<String, String> peerAddresses) {
 680             this.shardName = shardName;
 681             this.shardId = shardId;
 682             this.peerAddresses = peerAddresses;
 683         }
 684
 685         String getShardName() {
 686             return shardName;
 687         }
 688
 689         ActorRef getActor(){
 690             return actor;
 691         }
 692
 693         ActorPath getActorPath() {
 694             return actorPath;
 695         }
 696
 697         void setActor(ActorRef actor) {
 698             this.actor = actor;
 699             this.actorPath = actor.path();
 700         }
 701
 702         ShardIdentifier getShardId() {
 703             return shardId;
 704         }
 705
 706         Map<String, String> getPeerAddresses() {
 707             return peerAddresses;
 708         }
 709
 710         void updatePeerAddress(String peerId, String peerAddress, ActorRef sender){
 711             LOG.info("updatePeerAddress for peer {} with address {}", peerId,
 712                 peerAddress);
 713             if(peerAddresses.containsKey(peerId)){
 714                 peerAddresses.put(peerId, peerAddress);
 715
 716                 if(actor != null) {
 717                     if(LOG.isDebugEnabled()) {
 718                         LOG.debug("Sending PeerAddressResolved for peer {} with address {} to {}",
 719                                 peerId, peerAddress, actor.path());
 720                     }
 721
 722                     actor.tell(new PeerAddressResolved(peerId.toString(), peerAddress), sender);
 723                 }
 724
 725                 notifyOnShardInitializedCallbacks();
 726             }
 727         }
 728
 729         boolean isShardReady() {
 730             return !RaftState.Candidate.name().equals(role) && !Strings.isNullOrEmpty(role);
 731         }
 732
 733         boolean isShardReadyWithLeaderId() {
 734             return isShardReady() && (isLeader() || peerAddresses.containsKey(leaderId));
 735         }
 736
 737         boolean isShardInitialized() {
 738             return getActor() != null && actorInitialized;
 739         }
 740
 741         boolean isLeader() {
 742             return Objects.equal(leaderId, shardId.toString());
 743         }
 744
 745         String getSerializedLeaderActor() {
 746             if(isLeader()) {
 747                 return Serialization.serializedActorPath(getActor());
 748             } else {
 749                 return peerAddresses.get(leaderId);
 750             }
 751         }
 752
 753         void setActorInitialized() {
 754             LOG.debug("Shard {} is initialized", shardId);
 755
 756             this.actorInitialized = true;
 757
 758             notifyOnShardInitializedCallbacks();
 759         }
 760
 761         private void notifyOnShardInitializedCallbacks() {
 762             if(onShardInitializedSet.isEmpty()) {
 763                 return;
 764             }
 765
 766             boolean ready = isShardReadyWithLeaderId();
 767
 768             if(LOG.isDebugEnabled()) {
 769                 LOG.debug("Shard {} is {} - notifying {} OnShardInitialized callbacks", shardId,
 770                         ready ? "ready" : "initialized", onShardInitializedSet.size());
 771             }
 772
 773             Iterator<OnShardInitialized> iter = onShardInitializedSet.iterator();
 774             while(iter.hasNext()) {
 775                 OnShardInitialized onShardInitialized = iter.next();
 776                 if(!(onShardInitialized instanceof OnShardReady) || ready) {
 777                     iter.remove();
 778                     onShardInitialized.getTimeoutSchedule().cancel();
 779                     onShardInitialized.getReplyRunnable().run();
 780                 }
 781             }
 782         }
 783
 784         void addOnShardInitialized(OnShardInitialized onShardInitialized) {
 785             onShardInitializedSet.add(onShardInitialized);
 786         }
 787
 788         void removeOnShardInitialized(OnShardInitialized onShardInitialized) {
 789             onShardInitializedSet.remove(onShardInitialized);
 790         }
 791
 792         void setRole(String newRole) {
 793             this.role = newRole;
 794
 795             notifyOnShardInitializedCallbacks();
 796         }
 797
 798         void setFollowerSyncStatus(boolean syncStatus){
 799             this.followerSyncStatus = syncStatus;
 800         }
 801
 802         boolean isInSync(){
 803             if(RaftState.Follower.name().equals(this.role)){
 804                 return followerSyncStatus;
 805             } else if(RaftState.Leader.name().equals(this.role)){
 806                 return true;
 807             }
 808
 809             return false;
 810         }
 811
 812         void setLeaderId(String leaderId) {
 813             this.leaderId = leaderId;
 814
 815             notifyOnShardInitializedCallbacks();
 816         }
 817     }
 818
 819     private static class ShardManagerCreator implements Creator<ShardManager> {
 820         private static final long serialVersionUID = 1L;
 821
 822         final ClusterWrapper cluster;
 823         final Configuration configuration;
 824         final DatastoreContext datastoreContext;
 825         private final CountDownLatch waitTillReadyCountdownLatch;
 826
 827         ShardManagerCreator(ClusterWrapper cluster,
 828                             Configuration configuration, DatastoreContext datastoreContext, CountDownLatch waitTillReadyCountdownLatch) {
 829             this.cluster = cluster;
 830             this.configuration = configuration;
 831             this.datastoreContext = datastoreContext;
 832             this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
 833         }
 834
 835         @Override
 836         public ShardManager create() throws Exception {
 837             return new ShardManager(cluster, configuration, datastoreContext, waitTillReadyCountdownLatch);
 838         }
 839     }
 840
 841     private static class OnShardInitialized {
 842         private final Runnable replyRunnable;
 843         private Cancellable timeoutSchedule;
 844
 845         OnShardInitialized(Runnable replyRunnable) {
 846             this.replyRunnable = replyRunnable;
 847         }
 848
 849         Runnable getReplyRunnable() {
 850             return replyRunnable;
 851         }
 852
 853         Cancellable getTimeoutSchedule() {
 854             return timeoutSchedule;
 855         }
 856
 857         void setTimeoutSchedule(Cancellable timeoutSchedule) {
 858             this.timeoutSchedule = timeoutSchedule;
 859         }
 860     }
 861
 862     private static class OnShardReady extends OnShardInitialized {
 863         OnShardReady(Runnable replyRunnable) {
 864             super(replyRunnable);
 865         }
 866     }
 867
 868     private static class ShardNotInitializedTimeout {
 869         private final ActorRef sender;
 870         private final ShardInformation shardInfo;
 871         private final OnShardInitialized onShardInitialized;
 872
 873         ShardNotInitializedTimeout(ShardInformation shardInfo, OnShardInitialized onShardInitialized, ActorRef sender) {
 874             this.sender = sender;
 875             this.shardInfo = shardInfo;
 876             this.onShardInitialized = onShardInitialized;
 877         }
 878
 879         ActorRef getSender() {
 880             return sender;
 881         }
 882
 883         ShardInformation getShardInfo() {
 884             return shardInfo;
 885         }
 886
 887         OnShardInitialized getOnShardInitialized() {
 888             return onShardInitialized;
 889         }
 890     }
 891
 892     static class SchemaContextModules implements Serializable {
 893         private static final long serialVersionUID = -8884620101025936590L;
 894
 895         private final Set<String> modules;
 896
 897         SchemaContextModules(Set<String> modules){
 898             this.modules = modules;
 899         }
 900
 901         public Set<String> getModules() {
 902             return modules;
 903         }
 904     }
 905 }
 906
 907
 908