opendaylight/md-sal/sal-akka-raft/src/main/java/org/opendaylight/controller/cluster/raft/behaviors/AbstractLeader.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.raft.behaviors;
  10
  11 import akka.actor.ActorRef;
  12 import akka.actor.ActorSelection;
  13 import akka.actor.Cancellable;
  14 import com.google.common.annotations.VisibleForTesting;
  15 import com.google.common.base.Optional;
  16 import com.google.common.base.Preconditions;
  17 import com.google.protobuf.ByteString;
  18 import java.io.IOException;
  19 import java.util.ArrayList;
  20 import java.util.Collections;
  21 import java.util.HashMap;
  22 import java.util.List;
  23 import java.util.Map;
  24 import java.util.Set;
  25 import java.util.concurrent.TimeUnit;
  26 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  27 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  28 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  29 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  30 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  31 import org.opendaylight.controller.cluster.raft.RaftState;
  32 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  33 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  34 import org.opendaylight.controller.cluster.raft.base.messages.InitiateInstallSnapshot;
  35 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  36 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  37 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  38 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  39 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  40 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  41 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  42 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  43 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  44 import scala.concurrent.duration.FiniteDuration;
  45
  46 /**
  47  * The behavior of a RaftActor when it is in the Leader state
  48  * <p/>
  49  * Leaders:
  50  * <ul>
  51  * <li> Upon election: send initial empty AppendEntries RPCs
  52  * (heartbeat) to each server; repeat during idle periods to
  53  * prevent election timeouts (§5.2)
  54  * <li> If command received from client: append entry to local log,
  55  * respond after entry applied to state machine (§5.3)
  56  * <li> If last log index ≥ nextIndex for a follower: send
  57  * AppendEntries RPC with log entries starting at nextIndex
  58  * <ul>
  59  * <li> If successful: update nextIndex and matchIndex for
  60  * follower (§5.3)
  61  * <li> If AppendEntries fails because of log inconsistency:
  62  * decrement nextIndex and retry (§5.3)
  63  * </ul>
  64  * <li> If there exists an N such that N > commitIndex, a majority
  65  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  66  * set commitIndex = N (§5.3, §5.4).
  67  */
  68 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
  69
  70     // The index of the first chunk that is sent when installing a snapshot
  71     public static final int FIRST_CHUNK_INDEX = 1;
  72
  73     // The index that the follower should respond with if it needs the install snapshot to be reset
  74     public static final int INVALID_CHUNK_INDEX = -1;
  75
  76     // This would be passed as the hash code of the last chunk when sending the first chunk
  77     public static final int INITIAL_LAST_CHUNK_HASH_CODE = -1;
  78
  79     protected final Map<String, FollowerLogInformation> followerToLog = new HashMap<>();
  80     protected final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
  81
  82     protected final Set<String> followers;
  83
  84     private Cancellable heartbeatSchedule = null;
  85
  86     private List<ClientRequestTracker> trackerList = new ArrayList<>();
  87
  88     protected final int minReplicationCount;
  89
  90     protected final int minIsolatedLeaderPeerCount;
  91
  92     private Optional<ByteString> snapshot;
  93
  94     public AbstractLeader(RaftActorContext context) {
  95         super(context);
  96
  97         followers = context.getPeerAddresses().keySet();
  98
  99         for (String followerId : followers) {
 100             FollowerLogInformation followerLogInformation =
 101                 new FollowerLogInformationImpl(followerId,
 102                     context.getCommitIndex(), -1,
 103                     context.getConfigParams().getElectionTimeOutInterval());
 104
 105             followerToLog.put(followerId, followerLogInformation);
 106         }
 107
 108         leaderId = context.getId();
 109
 110         if(LOG.isDebugEnabled()) {
 111             LOG.debug("Election:Leader has following peers: {}", followers);
 112         }
 113
 114         minReplicationCount = getMajorityVoteCount(followers.size());
 115
 116         // the isolated Leader peer count will be 1 less than the majority vote count.
 117         // this is because the vote count has the self vote counted in it
 118         // for e.g
 119         // 0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
 120         // 2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
 121         // 4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
 122         minIsolatedLeaderPeerCount = minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
 123
 124         snapshot = Optional.absent();
 125
 126         // Immediately schedule a heartbeat
 127         // Upon election: send initial empty AppendEntries RPCs
 128         // (heartbeat) to each server; repeat during idle periods to
 129         // prevent election timeouts (§5.2)
 130         scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
 131     }
 132
 133     private Optional<ByteString> getSnapshot() {
 134         return snapshot;
 135     }
 136
 137     @VisibleForTesting
 138     void setSnapshot(Optional<ByteString> snapshot) {
 139         this.snapshot = snapshot;
 140     }
 141
 142     @Override
 143     protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 144         AppendEntries appendEntries) {
 145
 146         if(LOG.isDebugEnabled()) {
 147             LOG.debug(appendEntries.toString());
 148         }
 149
 150         return this;
 151     }
 152
 153     @Override
 154     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 155         AppendEntriesReply appendEntriesReply) {
 156
 157         if(! appendEntriesReply.isSuccess()) {
 158             if(LOG.isDebugEnabled()) {
 159                 LOG.debug(appendEntriesReply.toString());
 160             }
 161         }
 162
 163         // Update the FollowerLogInformation
 164         String followerId = appendEntriesReply.getFollowerId();
 165         FollowerLogInformation followerLogInformation =
 166             followerToLog.get(followerId);
 167
 168         if(followerLogInformation == null){
 169             LOG.error("Unknown follower {}", followerId);
 170             return this;
 171         }
 172
 173         followerLogInformation.markFollowerActive();
 174
 175         if (appendEntriesReply.isSuccess()) {
 176             followerLogInformation
 177                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 178             followerLogInformation
 179                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 180         } else {
 181
 182             // TODO: When we find that the follower is out of sync with the
 183             // Leader we simply decrement that followers next index by 1.
 184             // Would it be possible to do better than this? The RAFT spec
 185             // does not explicitly deal with it but may be something for us to
 186             // think about
 187
 188             followerLogInformation.decrNextIndex();
 189         }
 190
 191         // Now figure out if this reply warrants a change in the commitIndex
 192         // If there exists an N such that N > commitIndex, a majority
 193         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 194         // set commitIndex = N (§5.3, §5.4).
 195         for (long N = context.getCommitIndex() + 1; ; N++) {
 196             int replicatedCount = 1;
 197
 198             for (FollowerLogInformation info : followerToLog.values()) {
 199                 if (info.getMatchIndex().get() >= N) {
 200                     replicatedCount++;
 201                 }
 202             }
 203
 204             if (replicatedCount >= minReplicationCount) {
 205                 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
 206                 if (replicatedLogEntry != null &&
 207                     replicatedLogEntry.getTerm() == currentTerm()) {
 208                     context.setCommitIndex(N);
 209                 }
 210             } else {
 211                 break;
 212             }
 213         }
 214
 215         // Apply the change to the state machine
 216         if (context.getCommitIndex() > context.getLastApplied()) {
 217             applyLogToStateMachine(context.getCommitIndex());
 218         }
 219
 220         return this;
 221     }
 222
 223     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 224
 225         ClientRequestTracker toRemove = findClientRequestTracker(logIndex);
 226         if(toRemove != null) {
 227             trackerList.remove(toRemove);
 228         }
 229
 230         return toRemove;
 231     }
 232
 233     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 234         for (ClientRequestTracker tracker : trackerList) {
 235             if (tracker.getIndex() == logIndex) {
 236                 return tracker;
 237             }
 238         }
 239         return null;
 240     }
 241
 242     @Override
 243     protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 244         RequestVoteReply requestVoteReply) {
 245         return this;
 246     }
 247
 248     @Override
 249     public RaftState state() {
 250         return RaftState.Leader;
 251     }
 252
 253     @Override
 254     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 255         Preconditions.checkNotNull(sender, "sender should not be null");
 256
 257         Object message = fromSerializableMessage(originalMessage);
 258
 259         if (message instanceof RaftRPC) {
 260             RaftRPC rpc = (RaftRPC) message;
 261             // If RPC request or response contains term T > currentTerm:
 262             // set currentTerm = T, convert to follower (§5.1)
 263             // This applies to all RPC messages and responses
 264             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 265                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 266
 267                 return switchBehavior(new Follower(context));
 268             }
 269         }
 270
 271         try {
 272             if (message instanceof SendHeartBeat) {
 273                 sendHeartBeat();
 274                 return this;
 275
 276             } else if(message instanceof InitiateInstallSnapshot) {
 277                 installSnapshotIfNeeded();
 278
 279             } else if(message instanceof SendInstallSnapshot) {
 280                 // received from RaftActor
 281                 setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
 282                 sendInstallSnapshot();
 283
 284             } else if (message instanceof Replicate) {
 285                 replicate((Replicate) message);
 286
 287             } else if (message instanceof InstallSnapshotReply){
 288                 handleInstallSnapshotReply((InstallSnapshotReply) message);
 289
 290             }
 291         } finally {
 292             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 293         }
 294
 295         return super.handleMessage(sender, message);
 296     }
 297
 298     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 299         String followerId = reply.getFollowerId();
 300         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 301         FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 302         followerLogInformation.markFollowerActive();
 303
 304         if (followerToSnapshot != null &&
 305             followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 306
 307             if (reply.isSuccess()) {
 308                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 309                     //this was the last chunk reply
 310                     if(LOG.isDebugEnabled()) {
 311                         LOG.debug("InstallSnapshotReply received, " +
 312                                 "last chunk received, Chunk:{}. Follower:{} Setting nextIndex:{}",
 313                             reply.getChunkIndex(), followerId,
 314                             context.getReplicatedLog().getSnapshotIndex() + 1
 315                         );
 316                     }
 317
 318                     followerLogInformation.setMatchIndex(
 319                         context.getReplicatedLog().getSnapshotIndex());
 320                     followerLogInformation.setNextIndex(
 321                         context.getReplicatedLog().getSnapshotIndex() + 1);
 322                     mapFollowerToSnapshot.remove(followerId);
 323
 324                     if(LOG.isDebugEnabled()) {
 325                         LOG.debug("followerToLog.get(followerId).getNextIndex().get()=" +
 326                             followerToLog.get(followerId).getNextIndex().get());
 327                     }
 328
 329                     if (mapFollowerToSnapshot.isEmpty()) {
 330                         // once there are no pending followers receiving snapshots
 331                         // we can remove snapshot from the memory
 332                         setSnapshot(Optional.<ByteString>absent());
 333                     }
 334
 335                 } else {
 336                     followerToSnapshot.markSendStatus(true);
 337                 }
 338             } else {
 339                 LOG.info("InstallSnapshotReply received, " +
 340                         "sending snapshot chunk failed, Will retry, Chunk:{}",
 341                     reply.getChunkIndex()
 342                 );
 343
 344                 followerToSnapshot.markSendStatus(false);
 345             }
 346
 347         } else {
 348             LOG.error("ERROR!!" +
 349                     "FollowerId in InstallSnapshotReply not known to Leader" +
 350                     " or Chunk Index in InstallSnapshotReply not matching {} != {}",
 351                 followerToSnapshot.getChunkIndex(), reply.getChunkIndex()
 352             );
 353
 354             if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
 355                 // Since the Follower did not find this index to be valid we should reset the follower snapshot
 356                 // so that Installing the snapshot can resume from the beginning
 357                 followerToSnapshot.reset();
 358             }
 359         }
 360     }
 361
 362     private void replicate(Replicate replicate) {
 363         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 364
 365         if(LOG.isDebugEnabled()) {
 366             LOG.debug("Replicate message {}", logIndex);
 367         }
 368
 369         // Create a tracker entry we will use this later to notify the
 370         // client actor
 371         trackerList.add(
 372             new ClientRequestTrackerImpl(replicate.getClientActor(),
 373                 replicate.getIdentifier(),
 374                 logIndex)
 375         );
 376
 377         if (followers.size() == 0) {
 378             context.setCommitIndex(logIndex);
 379             applyLogToStateMachine(logIndex);
 380         } else {
 381             sendAppendEntries();
 382         }
 383     }
 384
 385     private void sendAppendEntries() {
 386         // Send an AppendEntries to all followers
 387         for (String followerId : followers) {
 388             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 389
 390             if (followerActor != null) {
 391                 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 392                 long followerNextIndex = followerLogInformation.getNextIndex().get();
 393                 boolean isFollowerActive = followerLogInformation.isFollowerActive();
 394                 List<ReplicatedLogEntry> entries = null;
 395
 396                 if (mapFollowerToSnapshot.get(followerId) != null) {
 397                     // if install snapshot is in process , then sent next chunk if possible
 398                     if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 399                         sendSnapshotChunk(followerActor, followerId);
 400                     } else {
 401                         // we send a heartbeat even if we have not received a reply for the last chunk
 402                         sendAppendEntriesToFollower(followerActor, followerNextIndex,
 403                             Collections.<ReplicatedLogEntry>emptyList());
 404                     }
 405
 406                 } else {
 407                     long leaderLastIndex = context.getReplicatedLog().lastIndex();
 408                     long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 409
 410                     if (isFollowerActive &&
 411                         context.getReplicatedLog().isPresent(followerNextIndex)) {
 412                         // FIXME : Sending one entry at a time
 413                         entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
 414
 415                     } else if (isFollowerActive && followerNextIndex >= 0 &&
 416                         leaderLastIndex >= followerNextIndex ) {
 417                         // if the followers next index is not present in the leaders log, and
 418                         // if the follower is just not starting and if leader's index is more than followers index
 419                         // then snapshot should be sent
 420
 421                         if(LOG.isDebugEnabled()) {
 422                             LOG.debug("InitiateInstallSnapshot to follower:{}," +
 423                                     "follower-nextIndex:{}, leader-snapshot-index:{},  " +
 424                                     "leader-last-index:{}", followerId,
 425                                 followerNextIndex, leaderSnapShotIndex, leaderLastIndex
 426                             );
 427                         }
 428                         actor().tell(new InitiateInstallSnapshot(), actor());
 429
 430                         // we would want to sent AE as the capture snapshot might take time
 431                         entries =  Collections.<ReplicatedLogEntry>emptyList();
 432
 433                     } else {
 434                         //we send an AppendEntries, even if the follower is inactive
 435                         // in-order to update the followers timestamp, in case it becomes active again
 436                         entries =  Collections.<ReplicatedLogEntry>emptyList();
 437                     }
 438
 439                     sendAppendEntriesToFollower(followerActor, followerNextIndex, entries);
 440
 441                 }
 442             }
 443         }
 444     }
 445
 446     private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
 447         List<ReplicatedLogEntry> entries) {
 448         followerActor.tell(
 449             new AppendEntries(currentTerm(), context.getId(),
 450                 prevLogIndex(followerNextIndex),
 451                 prevLogTerm(followerNextIndex), entries,
 452                 context.getCommitIndex()).toSerializable(),
 453             actor()
 454         );
 455     }
 456
 457     /**
 458      * An installSnapshot is scheduled at a interval that is a multiple of
 459      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 460      * snapshots at every heartbeat.
 461      *
 462      * Install Snapshot works as follows
 463      * 1. Leader sends a InitiateInstallSnapshot message to self
 464      * 2. Leader then initiates the capture snapshot by sending a CaptureSnapshot message to actor
 465      * 3. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
 466      * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
 467      * 4. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
 468      * 5. On complete, Follower sends back a InstallSnapshotReply.
 469      * 6. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
 470      * and replenishes the memory by deleting the snapshot in Replicated log.
 471      *
 472      */
 473     private void installSnapshotIfNeeded() {
 474         for (String followerId : followers) {
 475             ActorSelection followerActor =
 476                 context.getPeerActorSelection(followerId);
 477
 478             if(followerActor != null) {
 479                 FollowerLogInformation followerLogInformation =
 480                     followerToLog.get(followerId);
 481
 482                 long nextIndex = followerLogInformation.getNextIndex().get();
 483
 484                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 485                     context.getReplicatedLog().isInSnapshot(nextIndex)) {
 486                     LOG.info("{} follower needs a snapshot install", followerId);
 487                     if (snapshot.isPresent()) {
 488                         // if a snapshot is present in the memory, most likely another install is in progress
 489                         // no need to capture snapshot
 490                         sendSnapshotChunk(followerActor, followerId);
 491
 492                     } else {
 493                         initiateCaptureSnapshot();
 494                         //we just need 1 follower who would need snapshot to be installed.
 495                         // when we have the snapshot captured, we would again check (in SendInstallSnapshot)
 496                         // who needs an install and send to all who need
 497                         break;
 498                     }
 499
 500                 }
 501             }
 502         }
 503     }
 504
 505     // on every install snapshot, we try to capture the snapshot.
 506     // Once a capture is going on, another one issued will get ignored by RaftActor.
 507     private void initiateCaptureSnapshot() {
 508         LOG.info("Initiating Snapshot Capture to Install Snapshot, Leader:{}", getLeaderId());
 509         ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
 510         long lastAppliedIndex = -1;
 511         long lastAppliedTerm = -1;
 512
 513         if (lastAppliedEntry != null) {
 514             lastAppliedIndex = lastAppliedEntry.getIndex();
 515             lastAppliedTerm = lastAppliedEntry.getTerm();
 516         } else if (context.getReplicatedLog().getSnapshotIndex() > -1)  {
 517             lastAppliedIndex = context.getReplicatedLog().getSnapshotIndex();
 518             lastAppliedTerm = context.getReplicatedLog().getSnapshotTerm();
 519         }
 520
 521         boolean isInstallSnapshotInitiated = true;
 522         actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(),
 523                 lastAppliedIndex, lastAppliedTerm, isInstallSnapshotInitiated),
 524             actor());
 525     }
 526
 527
 528     private void sendInstallSnapshot() {
 529         for (String followerId : followers) {
 530             ActorSelection followerActor = context.getPeerActorSelection(followerId);
 531
 532             if(followerActor != null) {
 533                 FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 534                 long nextIndex = followerLogInformation.getNextIndex().get();
 535
 536                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 537                     context.getReplicatedLog().isInSnapshot(nextIndex)) {
 538                     sendSnapshotChunk(followerActor, followerId);
 539                 }
 540             }
 541         }
 542     }
 543
 544     /**
 545      *  Sends a snapshot chunk to a given follower
 546      *  InstallSnapshot should qualify as a heartbeat too.
 547      */
 548     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 549         try {
 550             if (snapshot.isPresent()) {
 551                 followerActor.tell(
 552                     new InstallSnapshot(currentTerm(), context.getId(),
 553                         context.getReplicatedLog().getSnapshotIndex(),
 554                         context.getReplicatedLog().getSnapshotTerm(),
 555                         getNextSnapshotChunk(followerId,snapshot.get()),
 556                         mapFollowerToSnapshot.get(followerId).incrementChunkIndex(),
 557                         mapFollowerToSnapshot.get(followerId).getTotalChunks(),
 558                         Optional.of(mapFollowerToSnapshot.get(followerId).getLastChunkHashCode())
 559                     ).toSerializable(),
 560                     actor()
 561                 );
 562                 LOG.info("InstallSnapshot sent to follower {}, Chunk: {}/{}",
 563                     followerActor.path(), mapFollowerToSnapshot.get(followerId).getChunkIndex(),
 564                     mapFollowerToSnapshot.get(followerId).getTotalChunks());
 565             }
 566         } catch (IOException e) {
 567             LOG.error(e, "InstallSnapshot failed for Leader.");
 568         }
 569     }
 570
 571     /**
 572      * Acccepts snaphot as ByteString, enters into map for future chunks
 573      * creates and return a ByteString chunk
 574      */
 575     private ByteString getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 576         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 577         if (followerToSnapshot == null) {
 578             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 579             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 580         }
 581         ByteString nextChunk = followerToSnapshot.getNextChunk();
 582         if (LOG.isDebugEnabled()) {
 583             LOG.debug("Leader's snapshot nextChunk size:{}", nextChunk.size());
 584         }
 585         return nextChunk;
 586     }
 587
 588     private void sendHeartBeat() {
 589         if (followers.size() > 0) {
 590             sendAppendEntries();
 591         }
 592     }
 593
 594     private void stopHeartBeat() {
 595         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 596             heartbeatSchedule.cancel();
 597         }
 598     }
 599
 600     private void scheduleHeartBeat(FiniteDuration interval) {
 601         if(followers.size() == 0){
 602             // Optimization - do not bother scheduling a heartbeat as there are
 603             // no followers
 604             return;
 605         }
 606
 607         stopHeartBeat();
 608
 609         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 610         // message is sent to itself.
 611         // Scheduling the heartbeat only once here because heartbeats do not
 612         // need to be sent if there are other messages being sent to the remote
 613         // actor.
 614         heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
 615             interval, context.getActor(), new SendHeartBeat(),
 616             context.getActorSystem().dispatcher(), context.getActor());
 617     }
 618
 619     @Override
 620     public void close() throws Exception {
 621         stopHeartBeat();
 622     }
 623
 624     @Override
 625     public String getLeaderId() {
 626         return context.getId();
 627     }
 628
 629     protected boolean isLeaderIsolated() {
 630         int minPresent = minIsolatedLeaderPeerCount;
 631         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 632             if (followerLogInformation.isFollowerActive()) {
 633                 --minPresent;
 634                 if (minPresent == 0) {
 635                     break;
 636                 }
 637             }
 638         }
 639         return (minPresent != 0);
 640     }
 641
 642     /**
 643      * Encapsulates the snapshot bytestring and handles the logic of sending
 644      * snapshot chunks
 645      */
 646     protected class FollowerToSnapshot {
 647         private ByteString snapshotBytes;
 648         private int offset = 0;
 649         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 650         private int replyReceivedForOffset;
 651         // if replyStatus is false, the previous chunk is attempted
 652         private boolean replyStatus = false;
 653         private int chunkIndex;
 654         private int totalChunks;
 655         private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 656         private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 657
 658         public FollowerToSnapshot(ByteString snapshotBytes) {
 659             this.snapshotBytes = snapshotBytes;
 660             int size = snapshotBytes.size();
 661             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 662                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 663             if(LOG.isDebugEnabled()) {
 664                 LOG.debug("Snapshot {} bytes, total chunks to send:{}",
 665                     size, totalChunks);
 666             }
 667             replyReceivedForOffset = -1;
 668             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 669         }
 670
 671         public ByteString getSnapshotBytes() {
 672             return snapshotBytes;
 673         }
 674
 675         public int incrementOffset() {
 676             if(replyStatus) {
 677                 // if prev chunk failed, we would want to sent the same chunk again
 678                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 679             }
 680             return offset;
 681         }
 682
 683         public int incrementChunkIndex() {
 684             if (replyStatus) {
 685                 // if prev chunk failed, we would want to sent the same chunk again
 686                 chunkIndex =  chunkIndex + 1;
 687             }
 688             return chunkIndex;
 689         }
 690
 691         public int getChunkIndex() {
 692             return chunkIndex;
 693         }
 694
 695         public int getTotalChunks() {
 696             return totalChunks;
 697         }
 698
 699         public boolean canSendNextChunk() {
 700             // we only send a false if a chunk is sent but we have not received a reply yet
 701             return replyReceivedForOffset == offset;
 702         }
 703
 704         public boolean isLastChunk(int chunkIndex) {
 705             return totalChunks == chunkIndex;
 706         }
 707
 708         public void markSendStatus(boolean success) {
 709             if (success) {
 710                 // if the chunk sent was successful
 711                 replyReceivedForOffset = offset;
 712                 replyStatus = true;
 713                 lastChunkHashCode = nextChunkHashCode;
 714             } else {
 715                 // if the chunk sent was failure
 716                 replyReceivedForOffset = offset;
 717                 replyStatus = false;
 718             }
 719         }
 720
 721         public ByteString getNextChunk() {
 722             int snapshotLength = getSnapshotBytes().size();
 723             int start = incrementOffset();
 724             int size = context.getConfigParams().getSnapshotChunkSize();
 725             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 726                 size = snapshotLength;
 727             } else {
 728                 if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 729                     size = snapshotLength - start;
 730                 }
 731             }
 732
 733             if(LOG.isDebugEnabled()) {
 734                 LOG.debug("length={}, offset={},size={}",
 735                     snapshotLength, start, size);
 736             }
 737             ByteString substring = getSnapshotBytes().substring(start, start + size);
 738             nextChunkHashCode = substring.hashCode();
 739             return substring;
 740         }
 741
 742         /**
 743          * reset should be called when the Follower needs to be sent the snapshot from the beginning
 744          */
 745         public void reset(){
 746             offset = 0;
 747             replyStatus = false;
 748             replyReceivedForOffset = offset;
 749             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 750             lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 751         }
 752
 753         public int getLastChunkHashCode() {
 754             return lastChunkHashCode;
 755         }
 756     }
 757
 758     // called from example-actor for printing the follower-states
 759     public String printFollowerStates() {
 760         final StringBuilder sb = new StringBuilder();
 761
 762         sb.append('[');
 763         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 764             sb.append('{');
 765             sb.append(followerLogInformation.getId());
 766             sb.append(" state:");
 767             sb.append(followerLogInformation.isFollowerActive());
 768             sb.append("},");
 769         }
 770         sb.append(']');
 771
 772         return sb.toString();
 773     }
 774
 775     @VisibleForTesting
 776     void markFollowerActive(String followerId) {
 777         followerToLog.get(followerId).markFollowerActive();
 778     }
 779 }