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