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.jmx.mbeans.shardmanager.ShardManagerInfo;
  48 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfoMBean;
  49 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
  50 import org.opendaylight.controller.cluster.datastore.messages.CreateShard;
  51 import org.opendaylight.controller.cluster.datastore.messages.CreateShardReply;
  52 import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard;
  53 import org.opendaylight.controller.cluster.datastore.messages.FindPrimary;
  54 import org.opendaylight.controller.cluster.datastore.messages.LocalPrimaryShardFound;
  55 import org.opendaylight.controller.cluster.datastore.messages.LocalShardFound;
  56 import org.opendaylight.controller.cluster.datastore.messages.LocalShardNotFound;
  57 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
  58 import org.opendaylight.controller.cluster.datastore.messages.PeerDown;
  59 import org.opendaylight.controller.cluster.datastore.messages.PeerUp;
  60 import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary;
  61 import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound;
  62 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
  63 import org.opendaylight.controller.cluster.datastore.messages.SwitchShardBehavior;
  64 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
  65 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
  66 import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache;
  67 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
  68 import org.opendaylight.controller.cluster.notifications.RoleChangeNotification;
  69 import org.opendaylight.controller.cluster.raft.RaftState;
  70 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
  71 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
  73 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  74 import org.slf4j.Logger;
  75 import org.slf4j.LoggerFactory;
  76 import scala.concurrent.duration.Duration;
  77 import scala.concurrent.duration.FiniteDuration;
  78
  79 /**
  80  * The ShardManager has the following jobs,
  81  * <ul>
  82  * <li> Create all the local shard replicas that belong on this cluster member
  83  * <li> Find the address of the local shard
  84  * <li> Find the primary replica for any given shard
  85  * <li> Monitor the cluster members and store their addresses
  86  * <ul>
  87  */
  88 public class ShardManager extends AbstractUntypedPersistentActorWithMetering {
  89
  90     private static final Logger LOG = LoggerFactory.getLogger(ShardManager.class);
  91
  92     // Stores a mapping between a 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 ClusterWrapper cluster;
 102
 103     private final Configuration configuration;
 104
 105     private final String shardDispatcherPath;
 106
 107     private ShardManagerInfo mBean;
 108
 109     private DatastoreContext datastoreContext;
 110
 111     private final CountDownLatch waitTillReadyCountdownLatch;
 112
 113     private final PrimaryShardInfoFutureCache primaryShardInfoCache;
 114
 115     private final ShardPeerAddressResolver peerAddressResolver;
 116
 117     private SchemaContext schemaContext;
 118
 119     /**
 120      */
 121     protected ShardManager(ClusterWrapper cluster, Configuration configuration,
 122             DatastoreContext datastoreContext, CountDownLatch waitTillReadyCountdownLatch,
 123             PrimaryShardInfoFutureCache primaryShardInfoCache) {
 124
 125         this.cluster = Preconditions.checkNotNull(cluster, "cluster should not be null");
 126         this.configuration = Preconditions.checkNotNull(configuration, "configuration should not be null");
 127         this.datastoreContext = datastoreContext;
 128         this.type = datastoreContext.getDataStoreType();
 129         this.shardDispatcherPath =
 130                 new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Shard);
 131         this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
 132         this.primaryShardInfoCache = primaryShardInfoCache;
 133
 134         peerAddressResolver = new ShardPeerAddressResolver(type, cluster.getCurrentMemberName());
 135         this.datastoreContext = DatastoreContext.newBuilderFrom(datastoreContext).shardPeerAddressResolver(
 136                 peerAddressResolver).build();
 137
 138         // Subscribe this actor to cluster member events
 139         cluster.subscribeToMemberEvents(getSelf());
 140
 141         createLocalShards();
 142     }
 143
 144     public static Props props(
 145             final ClusterWrapper cluster,
 146             final Configuration configuration,
 147             final DatastoreContext datastoreContext,
 148             final CountDownLatch waitTillReadyCountdownLatch,
 149             final PrimaryShardInfoFutureCache primaryShardInfoCache) {
 150
 151         Preconditions.checkNotNull(cluster, "cluster should not be null");
 152         Preconditions.checkNotNull(configuration, "configuration should not be null");
 153         Preconditions.checkNotNull(waitTillReadyCountdownLatch, "waitTillReadyCountdownLatch should not be null");
 154         Preconditions.checkNotNull(primaryShardInfoCache, "primaryShardInfoCache should not be null");
 155
 156         return Props.create(new ShardManagerCreator(cluster, configuration, datastoreContext,
 157                 waitTillReadyCountdownLatch, primaryShardInfoCache));
 158     }
 159
 160     @Override
 161     public void postStop() {
 162         LOG.info("Stopping ShardManager");
 163
 164         mBean.unregisterMBean();
 165     }
 166
 167     @Override
 168     public void handleCommand(Object message) throws Exception {
 169         if (message  instanceof FindPrimary) {
 170             findPrimary((FindPrimary)message);
 171         } else if(message instanceof FindLocalShard){
 172             findLocalShard((FindLocalShard) message);
 173         } else if (message instanceof UpdateSchemaContext) {
 174             updateSchemaContext(message);
 175         } else if(message instanceof ActorInitialized) {
 176             onActorInitialized(message);
 177         } else if (message instanceof ClusterEvent.MemberUp){
 178             memberUp((ClusterEvent.MemberUp) message);
 179         } else if (message instanceof ClusterEvent.MemberExited){
 180             memberExited((ClusterEvent.MemberExited) message);
 181         } else if(message instanceof ClusterEvent.MemberRemoved) {
 182             memberRemoved((ClusterEvent.MemberRemoved) message);
 183         } else if(message instanceof ClusterEvent.UnreachableMember) {
 184             memberUnreachable((ClusterEvent.UnreachableMember)message);
 185         } else if(message instanceof ClusterEvent.ReachableMember) {
 186             memberReachable((ClusterEvent.ReachableMember) message);
 187         } else if(message instanceof DatastoreContext) {
 188             onDatastoreContext((DatastoreContext)message);
 189         } else if(message instanceof RoleChangeNotification) {
 190             onRoleChangeNotification((RoleChangeNotification) message);
 191         } else if(message instanceof FollowerInitialSyncUpStatus){
 192             onFollowerInitialSyncStatus((FollowerInitialSyncUpStatus) message);
 193         } else if(message instanceof ShardNotInitializedTimeout) {
 194             onShardNotInitializedTimeout((ShardNotInitializedTimeout)message);
 195         } else if(message instanceof ShardLeaderStateChanged) {
 196             onLeaderStateChanged((ShardLeaderStateChanged) message);
 197         } else if(message instanceof SwitchShardBehavior){
 198             onSwitchShardBehavior((SwitchShardBehavior) message);
 199         } else if(message instanceof CreateShard) {
 200             onCreateShard((CreateShard)message);
 201         } else {
 202             unknownMessage(message);
 203         }
 204
 205     }
 206
 207     private void onCreateShard(CreateShard createShard) {
 208         Object reply;
 209         try {
 210             ModuleShardConfiguration moduleShardConfig = createShard.getModuleShardConfig();
 211             if(localShards.containsKey(moduleShardConfig.getShardName())) {
 212                 throw new IllegalStateException(String.format("Shard with name %s already exists",
 213                         moduleShardConfig.getShardName()));
 214             }
 215
 216             configuration.addModuleShardConfiguration(moduleShardConfig);
 217
 218             ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), moduleShardConfig.getShardName());
 219             Map<String, String> peerAddresses = getPeerAddresses(moduleShardConfig.getShardName()/*,
 220                     moduleShardConfig.getShardMemberNames()*/);
 221
 222             LOG.debug("onCreateShard: shardId: {}, memberNames: {}. peerAddresses: {}", shardId,
 223                     moduleShardConfig.getShardMemberNames(), peerAddresses);
 224
 225             DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
 226             if(shardDatastoreContext == null) {
 227                 shardDatastoreContext = datastoreContext;
 228             } else {
 229                 shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).shardPeerAddressResolver(
 230                         peerAddressResolver).build();
 231             }
 232
 233             ShardInformation info = new ShardInformation(moduleShardConfig.getShardName(), shardId, peerAddresses,
 234                     shardDatastoreContext, createShard.getShardPropsCreator(), peerAddressResolver);
 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         peerAddressResolver.removePeerAddress(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         peerAddressResolver.removePeerAddress(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         addPeerAddress(memberName, message.member().address());
 504
 505         checkReady();
 506     }
 507
 508     private void addPeerAddress(String memberName, Address address) {
 509         peerAddressResolver.addPeerAddress(memberName, address);
 510
 511         for(ShardInformation info : localShards.values()){
 512             String shardName = info.getShardName();
 513             String peerId = getShardIdentifier(memberName, shardName).toString();
 514             info.updatePeerAddress(peerId, peerAddressResolver.getShardActorAddress(shardName, memberName), getSelf());
 515
 516             info.peerUp(memberName, peerId, getSelf());
 517         }
 518     }
 519
 520     private void memberReachable(ClusterEvent.ReachableMember message) {
 521         String memberName = message.member().roles().head();
 522         LOG.debug("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
 523
 524         addPeerAddress(memberName, message.member().address());
 525
 526         markMemberAvailable(memberName);
 527     }
 528
 529     private void memberUnreachable(ClusterEvent.UnreachableMember message) {
 530         String memberName = message.member().roles().head();
 531         LOG.debug("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
 532
 533         markMemberUnavailable(memberName);
 534     }
 535
 536     private void markMemberUnavailable(final String memberName) {
 537         for(ShardInformation info : localShards.values()){
 538             String leaderId = info.getLeaderId();
 539             if(leaderId != null && leaderId.contains(memberName)) {
 540                 LOG.debug("Marking Leader {} as unavailable.", leaderId);
 541                 info.setLeaderAvailable(false);
 542
 543                 primaryShardInfoCache.remove(info.getShardName());
 544             }
 545
 546             info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 547         }
 548     }
 549
 550     private void markMemberAvailable(final String memberName) {
 551         for(ShardInformation info : localShards.values()){
 552             String leaderId = info.getLeaderId();
 553             if(leaderId != null && leaderId.contains(memberName)) {
 554                 LOG.debug("Marking Leader {} as available.", leaderId);
 555                 info.setLeaderAvailable(true);
 556             }
 557
 558             info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
 559         }
 560     }
 561
 562     private void onDatastoreContext(DatastoreContext context) {
 563         datastoreContext = DatastoreContext.newBuilderFrom(context).shardPeerAddressResolver(
 564                 peerAddressResolver).build();
 565         for (ShardInformation info : localShards.values()) {
 566             if (info.getActor() != null) {
 567                 info.getActor().tell(datastoreContext, getSelf());
 568             }
 569         }
 570     }
 571
 572     private void onSwitchShardBehavior(SwitchShardBehavior message) {
 573         ShardIdentifier identifier = ShardIdentifier.builder().fromShardIdString(message.getShardName()).build();
 574
 575         ShardInformation shardInformation = localShards.get(identifier.getShardName());
 576
 577         if(shardInformation != null && shardInformation.getActor() != null) {
 578             shardInformation.getActor().tell(
 579                     new SwitchBehavior(RaftState.valueOf(message.getNewState()), message.getTerm()), getSelf());
 580         } else {
 581             LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
 582                     message.getShardName(), message.getNewState());
 583         }
 584     }
 585
 586     /**
 587      * Notifies all the local shards of a change in the schema context
 588      *
 589      * @param message
 590      */
 591     private void updateSchemaContext(final Object message) {
 592         schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
 593
 594         LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getAllModuleIdentifiers().size());
 595
 596         for (ShardInformation info : localShards.values()) {
 597             if (info.getActor() == null) {
 598                 LOG.debug("Creating Shard {}", info.getShardId());
 599                 info.setActor(newShardActor(schemaContext, info));
 600             } else {
 601                 info.getActor().tell(message, getSelf());
 602             }
 603         }
 604     }
 605
 606     @VisibleForTesting
 607     protected ClusterWrapper getCluster() {
 608         return cluster;
 609     }
 610
 611     @VisibleForTesting
 612     protected ActorRef newShardActor(final SchemaContext schemaContext, ShardInformation info) {
 613         return getContext().actorOf(info.newProps(schemaContext)
 614                 .withDispatcher(shardDispatcherPath), info.getShardId().toString());
 615     }
 616
 617     private void findPrimary(FindPrimary message) {
 618         LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
 619
 620         final String shardName = message.getShardName();
 621         final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
 622
 623         // First see if the there is a local replica for the shard
 624         final ShardInformation info = localShards.get(shardName);
 625         if (info != null) {
 626             sendResponse(info, message.isWaitUntilReady(), true, new Supplier<Object>() {
 627                 @Override
 628                 public Object get() {
 629                     String primaryPath = info.getSerializedLeaderActor();
 630                     Object found = canReturnLocalShardState && info.isLeader() ?
 631                             new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
 632                                 new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
 633
 634                             if(LOG.isDebugEnabled()) {
 635                                 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
 636                             }
 637
 638                             return found;
 639                 }
 640             });
 641
 642             return;
 643         }
 644
 645         for(String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
 646             LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}", persistenceId(),
 647                     shardName, address);
 648
 649             getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
 650                     message.isWaitUntilReady()), getContext());
 651             return;
 652         }
 653
 654         LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
 655
 656         getSender().tell(new PrimaryNotFoundException(
 657                 String.format("No primary shard found for %s.", shardName)), getSelf());
 658     }
 659
 660     /**
 661      * Construct the name of the shard actor given the name of the member on
 662      * which the shard resides and the name of the shard
 663      *
 664      * @param memberName
 665      * @param shardName
 666      * @return
 667      */
 668     private ShardIdentifier getShardIdentifier(String memberName, String shardName){
 669         return peerAddressResolver.getShardIdentifier(memberName, shardName);
 670     }
 671
 672     /**
 673      * Create shards that are local to the member on which the ShardManager
 674      * runs
 675      *
 676      */
 677     private void createLocalShards() {
 678         String memberName = this.cluster.getCurrentMemberName();
 679         Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
 680
 681         ShardPropsCreator shardPropsCreator = new DefaultShardPropsCreator();
 682         List<String> localShardActorNames = new ArrayList<>();
 683         for(String shardName : memberShardNames){
 684             ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 685             Map<String, String> peerAddresses = getPeerAddresses(shardName);
 686             localShardActorNames.add(shardId.toString());
 687             localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses, datastoreContext,
 688                     shardPropsCreator, peerAddressResolver));
 689         }
 690
 691         mBean = ShardManagerInfo.createShardManagerMBean(memberName, "shard-manager-" + this.type,
 692                 datastoreContext.getDataStoreMXBeanType(), localShardActorNames);
 693
 694         mBean.setShardManager(this);
 695     }
 696
 697     /**
 698      * Given the name of the shard find the addresses of all it's peers
 699      *
 700      * @param shardName
 701      */
 702     private Map<String, String> getPeerAddresses(String shardName) {
 703         Collection<String> members = configuration.getMembersFromShardName(shardName);
 704         Map<String, String> peerAddresses = new HashMap<>();
 705
 706         String currentMemberName = this.cluster.getCurrentMemberName();
 707
 708         for(String memberName : members) {
 709             if(!currentMemberName.equals(memberName)) {
 710                 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
 711                 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
 712                 peerAddresses.put(shardId.toString(), address);
 713             }
 714         }
 715         return peerAddresses;
 716     }
 717
 718     @Override
 719     public SupervisorStrategy supervisorStrategy() {
 720
 721         return new OneForOneStrategy(10, Duration.create("1 minute"),
 722                 new Function<Throwable, SupervisorStrategy.Directive>() {
 723             @Override
 724             public SupervisorStrategy.Directive apply(Throwable t) {
 725                 LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
 726                 return SupervisorStrategy.resume();
 727             }
 728         }
 729                 );
 730
 731     }
 732
 733     @Override
 734     public String persistenceId() {
 735         return "shard-manager-" + type;
 736     }
 737
 738     @VisibleForTesting
 739     ShardManagerInfoMBean getMBean(){
 740         return mBean;
 741     }
 742
 743     @VisibleForTesting
 744     protected static class ShardInformation {
 745         private final ShardIdentifier shardId;
 746         private final String shardName;
 747         private ActorRef actor;
 748         private ActorPath actorPath;
 749         private final Map<String, String> initialPeerAddresses;
 750         private Optional<DataTree> localShardDataTree;
 751         private boolean leaderAvailable = false;
 752
 753         // flag that determines if the actor is ready for business
 754         private boolean actorInitialized = false;
 755
 756         private boolean followerSyncStatus = false;
 757
 758         private final Set<OnShardInitialized> onShardInitializedSet = Sets.newHashSet();
 759         private String role ;
 760         private String leaderId;
 761         private short leaderVersion;
 762
 763         private final DatastoreContext datastoreContext;
 764         private final ShardPropsCreator shardPropsCreator;
 765         private final ShardPeerAddressResolver addressResolver;
 766
 767         private ShardInformation(String shardName, ShardIdentifier shardId,
 768                 Map<String, String> initialPeerAddresses, DatastoreContext datastoreContext,
 769                 ShardPropsCreator shardPropsCreator, ShardPeerAddressResolver addressResolver) {
 770             this.shardName = shardName;
 771             this.shardId = shardId;
 772             this.initialPeerAddresses = initialPeerAddresses;
 773             this.datastoreContext = datastoreContext;
 774             this.shardPropsCreator = shardPropsCreator;
 775             this.addressResolver = addressResolver;
 776         }
 777
 778         Props newProps(SchemaContext schemaContext) {
 779             return shardPropsCreator.newProps(shardId, initialPeerAddresses, datastoreContext, schemaContext);
 780         }
 781
 782         String getShardName() {
 783             return shardName;
 784         }
 785
 786         ActorRef getActor(){
 787             return actor;
 788         }
 789
 790         ActorPath getActorPath() {
 791             return actorPath;
 792         }
 793
 794         void setActor(ActorRef actor) {
 795             this.actor = actor;
 796             this.actorPath = actor.path();
 797         }
 798
 799         ShardIdentifier getShardId() {
 800             return shardId;
 801         }
 802
 803         void setLocalDataTree(Optional<DataTree> localShardDataTree) {
 804             this.localShardDataTree = localShardDataTree;
 805         }
 806
 807         Optional<DataTree> getLocalShardDataTree() {
 808             return localShardDataTree;
 809         }
 810
 811         void updatePeerAddress(String peerId, String peerAddress, ActorRef sender){
 812             LOG.info("updatePeerAddress for peer {} with address {}", peerId, peerAddress);
 813
 814             if(actor != null) {
 815                 if(LOG.isDebugEnabled()) {
 816                     LOG.debug("Sending PeerAddressResolved for peer {} with address {} to {}",
 817                             peerId, peerAddress, actor.path());
 818                 }
 819
 820                 actor.tell(new PeerAddressResolved(peerId, peerAddress), sender);
 821             }
 822
 823             notifyOnShardInitializedCallbacks();
 824         }
 825
 826         void peerDown(String memberName, String peerId, ActorRef sender) {
 827             if(actor != null) {
 828                 actor.tell(new PeerDown(memberName, peerId), sender);
 829             }
 830         }
 831
 832         void peerUp(String memberName, String peerId, ActorRef sender) {
 833             if(actor != null) {
 834                 actor.tell(new PeerUp(memberName, peerId), sender);
 835             }
 836         }
 837
 838         boolean isShardReady() {
 839             return !RaftState.Candidate.name().equals(role) && !Strings.isNullOrEmpty(role);
 840         }
 841
 842         boolean isShardReadyWithLeaderId() {
 843             return leaderAvailable && isShardReady() && !RaftState.IsolatedLeader.name().equals(role) &&
 844                     (isLeader() || addressResolver.resolve(leaderId) != null);
 845         }
 846
 847         boolean isShardInitialized() {
 848             return getActor() != null && actorInitialized;
 849         }
 850
 851         boolean isLeader() {
 852             return Objects.equal(leaderId, shardId.toString());
 853         }
 854
 855         String getSerializedLeaderActor() {
 856             if(isLeader()) {
 857                 return Serialization.serializedActorPath(getActor());
 858             } else {
 859                 return addressResolver.resolve(leaderId);
 860             }
 861         }
 862
 863         void setActorInitialized() {
 864             LOG.debug("Shard {} is initialized", shardId);
 865
 866             this.actorInitialized = true;
 867
 868             notifyOnShardInitializedCallbacks();
 869         }
 870
 871         private void notifyOnShardInitializedCallbacks() {
 872             if(onShardInitializedSet.isEmpty()) {
 873                 return;
 874             }
 875
 876             boolean ready = isShardReadyWithLeaderId();
 877
 878             if(LOG.isDebugEnabled()) {
 879                 LOG.debug("Shard {} is {} - notifying {} OnShardInitialized callbacks", shardId,
 880                         ready ? "ready" : "initialized", onShardInitializedSet.size());
 881             }
 882
 883             Iterator<OnShardInitialized> iter = onShardInitializedSet.iterator();
 884             while(iter.hasNext()) {
 885                 OnShardInitialized onShardInitialized = iter.next();
 886                 if(!(onShardInitialized instanceof OnShardReady) || ready) {
 887                     iter.remove();
 888                     onShardInitialized.getTimeoutSchedule().cancel();
 889                     onShardInitialized.getReplyRunnable().run();
 890                 }
 891             }
 892         }
 893
 894         void addOnShardInitialized(OnShardInitialized onShardInitialized) {
 895             onShardInitializedSet.add(onShardInitialized);
 896         }
 897
 898         void removeOnShardInitialized(OnShardInitialized onShardInitialized) {
 899             onShardInitializedSet.remove(onShardInitialized);
 900         }
 901
 902         void setRole(String newRole) {
 903             this.role = newRole;
 904
 905             notifyOnShardInitializedCallbacks();
 906         }
 907
 908         void setFollowerSyncStatus(boolean syncStatus){
 909             this.followerSyncStatus = syncStatus;
 910         }
 911
 912         boolean isInSync(){
 913             if(RaftState.Follower.name().equals(this.role)){
 914                 return followerSyncStatus;
 915             } else if(RaftState.Leader.name().equals(this.role)){
 916                 return true;
 917             }
 918
 919             return false;
 920         }
 921
 922         boolean setLeaderId(String leaderId) {
 923             boolean changed = !Objects.equal(this.leaderId, leaderId);
 924             this.leaderId = leaderId;
 925             if(leaderId != null) {
 926                 this.leaderAvailable = true;
 927             }
 928             notifyOnShardInitializedCallbacks();
 929
 930             return changed;
 931         }
 932
 933         String getLeaderId() {
 934             return leaderId;
 935         }
 936
 937         void setLeaderAvailable(boolean leaderAvailable) {
 938             this.leaderAvailable = leaderAvailable;
 939         }
 940
 941         short getLeaderVersion() {
 942             return leaderVersion;
 943         }
 944
 945         void setLeaderVersion(short leaderVersion) {
 946             this.leaderVersion = leaderVersion;
 947         }
 948     }
 949
 950     private static class ShardManagerCreator implements Creator<ShardManager> {
 951         private static final long serialVersionUID = 1L;
 952
 953         final ClusterWrapper cluster;
 954         final Configuration configuration;
 955         final DatastoreContext datastoreContext;
 956         private final CountDownLatch waitTillReadyCountdownLatch;
 957         private final PrimaryShardInfoFutureCache primaryShardInfoCache;
 958
 959         ShardManagerCreator(ClusterWrapper cluster, Configuration configuration, DatastoreContext datastoreContext,
 960                 CountDownLatch waitTillReadyCountdownLatch, PrimaryShardInfoFutureCache primaryShardInfoCache) {
 961             this.cluster = cluster;
 962             this.configuration = configuration;
 963             this.datastoreContext = datastoreContext;
 964             this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
 965             this.primaryShardInfoCache = primaryShardInfoCache;
 966         }
 967
 968         @Override
 969         public ShardManager create() throws Exception {
 970             return new ShardManager(cluster, configuration, datastoreContext, waitTillReadyCountdownLatch,
 971                     primaryShardInfoCache);
 972         }
 973     }
 974
 975     private static class OnShardInitialized {
 976         private final Runnable replyRunnable;
 977         private Cancellable timeoutSchedule;
 978
 979         OnShardInitialized(Runnable replyRunnable) {
 980             this.replyRunnable = replyRunnable;
 981         }
 982
 983         Runnable getReplyRunnable() {
 984             return replyRunnable;
 985         }
 986
 987         Cancellable getTimeoutSchedule() {
 988             return timeoutSchedule;
 989         }
 990
 991         void setTimeoutSchedule(Cancellable timeoutSchedule) {
 992             this.timeoutSchedule = timeoutSchedule;
 993         }
 994     }
 995
 996     private static class OnShardReady extends OnShardInitialized {
 997         OnShardReady(Runnable replyRunnable) {
 998             super(replyRunnable);
 999         }
1000     }
1001
1002     private static class ShardNotInitializedTimeout {
1003         private final ActorRef sender;
1004         private final ShardInformation shardInfo;
1005         private final OnShardInitialized onShardInitialized;
1006
1007         ShardNotInitializedTimeout(ShardInformation shardInfo, OnShardInitialized onShardInitialized, ActorRef sender) {
1008             this.sender = sender;
1009             this.shardInfo = shardInfo;
1010             this.onShardInitialized = onShardInitialized;
1011         }
1012
1013         ActorRef getSender() {
1014             return sender;
1015         }
1016
1017         ShardInformation getShardInfo() {
1018             return shardInfo;
1019         }
1020
1021         OnShardInitialized getOnShardInitialized() {
1022             return onShardInitialized;
1023         }
1024     }
1025
1026     /**
1027      * We no longer persist SchemaContextModules but keep this class around for now for backwards
1028      * compatibility so we don't get de-serialization failures on upgrade from Helium.
1029      */
1030     @Deprecated
1031     static class SchemaContextModules implements Serializable {
1032         private static final long serialVersionUID = -8884620101025936590L;
1033
1034         private final Set<String> modules;
1035
1036         SchemaContextModules(Set<String> modules){
1037             this.modules = modules;
1038         }
1039
1040         public Set<String> getModules() {
1041             return modules;
1042         }
1043     }
1044 }
1045
1046
1047