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.common.collect.ImmutableMap;
  18 import com.google.common.collect.ImmutableMap.Builder;
  19 import com.google.protobuf.ByteString;
  20 import java.io.IOException;
  21 import java.util.Collection;
  22 import java.util.Collections;
  23 import java.util.HashMap;
  24 import java.util.Iterator;
  25 import java.util.LinkedList;
  26 import java.util.List;
  27 import java.util.Map;
  28 import java.util.Map.Entry;
  29 import java.util.concurrent.TimeUnit;
  30 import org.opendaylight.controller.cluster.raft.ClientRequestTracker;
  31 import org.opendaylight.controller.cluster.raft.ClientRequestTrackerImpl;
  32 import org.opendaylight.controller.cluster.raft.FollowerLogInformation;
  33 import org.opendaylight.controller.cluster.raft.FollowerLogInformationImpl;
  34 import org.opendaylight.controller.cluster.raft.RaftActorContext;
  35 import org.opendaylight.controller.cluster.raft.RaftState;
  36 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
  37 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshot;
  38 import org.opendaylight.controller.cluster.raft.base.messages.InitiateInstallSnapshot;
  39 import org.opendaylight.controller.cluster.raft.base.messages.Replicate;
  40 import org.opendaylight.controller.cluster.raft.base.messages.SendHeartBeat;
  41 import org.opendaylight.controller.cluster.raft.base.messages.SendInstallSnapshot;
  42 import org.opendaylight.controller.cluster.raft.messages.AppendEntries;
  43 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
  44 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshot;
  45 import org.opendaylight.controller.cluster.raft.messages.InstallSnapshotReply;
  46 import org.opendaylight.controller.cluster.raft.messages.RaftRPC;
  47 import org.opendaylight.controller.cluster.raft.messages.RequestVoteReply;
  48 import scala.concurrent.duration.FiniteDuration;
  49
  50 /**
  51  * The behavior of a RaftActor when it is in the Leader state
  52  * <p/>
  53  * Leaders:
  54  * <ul>
  55  * <li> Upon election: send initial empty AppendEntries RPCs
  56  * (heartbeat) to each server; repeat during idle periods to
  57  * prevent election timeouts (§5.2)
  58  * <li> If command received from client: append entry to local log,
  59  * respond after entry applied to state machine (§5.3)
  60  * <li> If last log index ≥ nextIndex for a follower: send
  61  * AppendEntries RPC with log entries starting at nextIndex
  62  * <ul>
  63  * <li> If successful: update nextIndex and matchIndex for
  64  * follower (§5.3)
  65  * <li> If AppendEntries fails because of log inconsistency:
  66  * decrement nextIndex and retry (§5.3)
  67  * </ul>
  68  * <li> If there exists an N such that N > commitIndex, a majority
  69  * of matchIndex[i] ≥ N, and log[N].term == currentTerm:
  70  * set commitIndex = N (§5.3, §5.4).
  71  */
  72 public abstract class AbstractLeader extends AbstractRaftActorBehavior {
  73
  74     // The index of the first chunk that is sent when installing a snapshot
  75     public static final int FIRST_CHUNK_INDEX = 1;
  76
  77     // The index that the follower should respond with if it needs the install snapshot to be reset
  78     public static final int INVALID_CHUNK_INDEX = -1;
  79
  80     // This would be passed as the hash code of the last chunk when sending the first chunk
  81     public static final int INITIAL_LAST_CHUNK_HASH_CODE = -1;
  82
  83     private final Map<String, FollowerLogInformation> followerToLog;
  84     private final Map<String, FollowerToSnapshot> mapFollowerToSnapshot = new HashMap<>();
  85
  86     private Cancellable heartbeatSchedule = null;
  87
  88     private final Collection<ClientRequestTracker> trackerList = new LinkedList<>();
  89
  90     protected final int minReplicationCount;
  91
  92     protected final int minIsolatedLeaderPeerCount;
  93
  94     private Optional<ByteString> snapshot;
  95
  96     private long replicatedToAllIndex = -1;
  97
  98     public AbstractLeader(RaftActorContext context) {
  99         super(context);
 100
 101         final Builder<String, FollowerLogInformation> ftlBuilder = ImmutableMap.builder();
 102         for (String followerId : context.getPeerAddresses().keySet()) {
 103             FollowerLogInformation followerLogInformation =
 104                 new FollowerLogInformationImpl(followerId,
 105                     context.getCommitIndex(), -1,
 106                     context.getConfigParams().getElectionTimeOutInterval());
 107
 108             ftlBuilder.put(followerId, followerLogInformation);
 109         }
 110         followerToLog = ftlBuilder.build();
 111
 112         leaderId = context.getId();
 113
 114         LOG.debug("{}: Election: Leader has following peers: {}", context.getId(), getFollowerIds());
 115
 116         minReplicationCount = getMajorityVoteCount(getFollowerIds().size());
 117
 118         // the isolated Leader peer count will be 1 less than the majority vote count.
 119         // this is because the vote count has the self vote counted in it
 120         // for e.g
 121         // 0 peers = 1 votesRequired , minIsolatedLeaderPeerCount = 0
 122         // 2 peers = 2 votesRequired , minIsolatedLeaderPeerCount = 1
 123         // 4 peers = 3 votesRequired, minIsolatedLeaderPeerCount = 2
 124         minIsolatedLeaderPeerCount = minReplicationCount > 0 ? (minReplicationCount - 1) : 0;
 125
 126         snapshot = Optional.absent();
 127
 128         // Immediately schedule a heartbeat
 129         // Upon election: send initial empty AppendEntries RPCs
 130         // (heartbeat) to each server; repeat during idle periods to
 131         // prevent election timeouts (§5.2)
 132         scheduleHeartBeat(new FiniteDuration(0, TimeUnit.SECONDS));
 133     }
 134
 135     /**
 136      * Return an immutable collection of follower identifiers.
 137      *
 138      * @return Collection of follower IDs
 139      */
 140     protected final Collection<String> getFollowerIds() {
 141         return followerToLog.keySet();
 142     }
 143
 144     private Optional<ByteString> getSnapshot() {
 145         return snapshot;
 146     }
 147
 148     @VisibleForTesting
 149     void setSnapshot(Optional<ByteString> snapshot) {
 150         this.snapshot = snapshot;
 151     }
 152
 153     @Override
 154     protected RaftActorBehavior handleAppendEntries(ActorRef sender,
 155         AppendEntries appendEntries) {
 156
 157         if(LOG.isDebugEnabled()) {
 158             LOG.debug("{}: handleAppendEntries: {}", context.getId(), appendEntries);
 159         }
 160
 161         return this;
 162     }
 163
 164     @Override
 165     protected RaftActorBehavior handleAppendEntriesReply(ActorRef sender,
 166         AppendEntriesReply appendEntriesReply) {
 167
 168         if(! appendEntriesReply.isSuccess()) {
 169             if(LOG.isDebugEnabled()) {
 170                 LOG.debug("{}: handleAppendEntriesReply: {}", context.getId(), appendEntriesReply);
 171             }
 172         }
 173
 174         // Update the FollowerLogInformation
 175         String followerId = appendEntriesReply.getFollowerId();
 176         FollowerLogInformation followerLogInformation =
 177             followerToLog.get(followerId);
 178
 179         if(followerLogInformation == null){
 180             LOG.error("{}: handleAppendEntriesReply - unknown follower {}", context.getId(), followerId);
 181             return this;
 182         }
 183
 184         followerLogInformation.markFollowerActive();
 185
 186         if (appendEntriesReply.isSuccess()) {
 187             followerLogInformation
 188                 .setMatchIndex(appendEntriesReply.getLogLastIndex());
 189             followerLogInformation
 190                 .setNextIndex(appendEntriesReply.getLogLastIndex() + 1);
 191         } else {
 192
 193             // TODO: When we find that the follower is out of sync with the
 194             // Leader we simply decrement that followers next index by 1.
 195             // Would it be possible to do better than this? The RAFT spec
 196             // does not explicitly deal with it but may be something for us to
 197             // think about
 198
 199             followerLogInformation.decrNextIndex();
 200         }
 201
 202         // Now figure out if this reply warrants a change in the commitIndex
 203         // If there exists an N such that N > commitIndex, a majority
 204         // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
 205         // set commitIndex = N (§5.3, §5.4).
 206         for (long N = context.getCommitIndex() + 1; ; N++) {
 207             int replicatedCount = 1;
 208
 209             for (FollowerLogInformation info : followerToLog.values()) {
 210                 if (info.getMatchIndex() >= N) {
 211                     replicatedCount++;
 212                 }
 213             }
 214
 215             if (replicatedCount >= minReplicationCount) {
 216                 ReplicatedLogEntry replicatedLogEntry = context.getReplicatedLog().get(N);
 217                 if (replicatedLogEntry != null &&
 218                     replicatedLogEntry.getTerm() == currentTerm()) {
 219                     context.setCommitIndex(N);
 220                 }
 221             } else {
 222                 break;
 223             }
 224         }
 225
 226         // Apply the change to the state machine
 227         if (context.getCommitIndex() > context.getLastApplied()) {
 228             applyLogToStateMachine(context.getCommitIndex());
 229         }
 230
 231         if (!context.isSnapshotCaptureInitiated()) {
 232             purgeInMemoryLog();
 233         }
 234
 235         //Send the next log entry immediately, if possible, no need to wait for heartbeat to trigger that event
 236         sendUpdatesToFollower(followerId, followerLogInformation, false);
 237         return this;
 238     }
 239
 240     private void purgeInMemoryLog() {
 241         //find the lowest index across followers which has been replicated to all. -1 if there are no followers.
 242         // we would delete the in-mem log from that index on, in-order to minimize mem usage
 243         // we would also share this info thru AE with the followers so that they can delete their log entries as well.
 244         long minReplicatedToAllIndex = followerToLog.isEmpty() ? -1 : Long.MAX_VALUE;
 245         for (FollowerLogInformation info : followerToLog.values()) {
 246             minReplicatedToAllIndex = Math.min(minReplicatedToAllIndex, info.getMatchIndex());
 247         }
 248
 249         replicatedToAllIndex = fakeSnapshot(minReplicatedToAllIndex, replicatedToAllIndex);
 250     }
 251
 252     @Override
 253     protected ClientRequestTracker removeClientRequestTracker(long logIndex) {
 254         final Iterator<ClientRequestTracker> it = trackerList.iterator();
 255         while (it.hasNext()) {
 256             final ClientRequestTracker t = it.next();
 257             if (t.getIndex() == logIndex) {
 258                 it.remove();
 259                 return t;
 260             }
 261         }
 262
 263         return null;
 264     }
 265
 266     @Override
 267     protected ClientRequestTracker findClientRequestTracker(long logIndex) {
 268         for (ClientRequestTracker tracker : trackerList) {
 269             if (tracker.getIndex() == logIndex) {
 270                 return tracker;
 271             }
 272         }
 273         return null;
 274     }
 275
 276     @Override
 277     protected RaftActorBehavior handleRequestVoteReply(ActorRef sender,
 278         RequestVoteReply requestVoteReply) {
 279         return this;
 280     }
 281
 282     @Override
 283     public RaftState state() {
 284         return RaftState.Leader;
 285     }
 286
 287     @Override
 288     public RaftActorBehavior handleMessage(ActorRef sender, Object originalMessage) {
 289         Preconditions.checkNotNull(sender, "sender should not be null");
 290
 291         Object message = fromSerializableMessage(originalMessage);
 292
 293         if (message instanceof RaftRPC) {
 294             RaftRPC rpc = (RaftRPC) message;
 295             // If RPC request or response contains term T > currentTerm:
 296             // set currentTerm = T, convert to follower (§5.1)
 297             // This applies to all RPC messages and responses
 298             if (rpc.getTerm() > context.getTermInformation().getCurrentTerm()) {
 299                 LOG.debug("{}: Term {} in \"{}\" message is greater than leader's term {}", context.getId(),
 300                         rpc.getTerm(), rpc, context.getTermInformation().getCurrentTerm());
 301
 302                 context.getTermInformation().updateAndPersist(rpc.getTerm(), null);
 303
 304                 return switchBehavior(new Follower(context));
 305             }
 306         }
 307
 308         try {
 309             if (message instanceof SendHeartBeat) {
 310                 sendHeartBeat();
 311                 return this;
 312
 313             } else if(message instanceof InitiateInstallSnapshot) {
 314                 installSnapshotIfNeeded();
 315
 316             } else if(message instanceof SendInstallSnapshot) {
 317                 // received from RaftActor
 318                 setSnapshot(Optional.of(((SendInstallSnapshot) message).getSnapshot()));
 319                 sendInstallSnapshot();
 320
 321             } else if (message instanceof Replicate) {
 322                 replicate((Replicate) message);
 323
 324             } else if (message instanceof InstallSnapshotReply){
 325                 handleInstallSnapshotReply((InstallSnapshotReply) message);
 326
 327             }
 328         } finally {
 329             scheduleHeartBeat(context.getConfigParams().getHeartBeatInterval());
 330         }
 331
 332         return super.handleMessage(sender, message);
 333     }
 334
 335     private void handleInstallSnapshotReply(InstallSnapshotReply reply) {
 336         String followerId = reply.getFollowerId();
 337         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 338
 339         if (followerToSnapshot == null) {
 340             LOG.error("{}: FollowerId {} in InstallSnapshotReply not known to Leader",
 341                     context.getId(), followerId);
 342             return;
 343         }
 344
 345         FollowerLogInformation followerLogInformation = followerToLog.get(followerId);
 346         followerLogInformation.markFollowerActive();
 347
 348         if (followerToSnapshot.getChunkIndex() == reply.getChunkIndex()) {
 349             boolean wasLastChunk = false;
 350             if (reply.isSuccess()) {
 351                 if(followerToSnapshot.isLastChunk(reply.getChunkIndex())) {
 352                     //this was the last chunk reply
 353                     if(LOG.isDebugEnabled()) {
 354                         LOG.debug("{}: InstallSnapshotReply received, " +
 355                                 "last chunk received, Chunk:{}. Follower:{} Setting nextIndex:{}",
 356                                 context.getId(), reply.getChunkIndex(), followerId,
 357                             context.getReplicatedLog().getSnapshotIndex() + 1
 358                         );
 359                     }
 360
 361                     followerLogInformation.setMatchIndex(
 362                         context.getReplicatedLog().getSnapshotIndex());
 363                     followerLogInformation.setNextIndex(
 364                         context.getReplicatedLog().getSnapshotIndex() + 1);
 365                     mapFollowerToSnapshot.remove(followerId);
 366
 367                     if(LOG.isDebugEnabled()) {
 368                         LOG.debug("{}: followerToLog.get(followerId).getNextIndex()=" +
 369                                 context.getId(), followerToLog.get(followerId).getNextIndex());
 370                     }
 371
 372                     if (mapFollowerToSnapshot.isEmpty()) {
 373                         // once there are no pending followers receiving snapshots
 374                         // we can remove snapshot from the memory
 375                         setSnapshot(Optional.<ByteString>absent());
 376                     }
 377                     wasLastChunk = true;
 378
 379                 } else {
 380                     followerToSnapshot.markSendStatus(true);
 381                 }
 382             } else {
 383                 LOG.info("{}: InstallSnapshotReply received sending snapshot chunk failed, Will retry, Chunk: {}",
 384                         context.getId(), reply.getChunkIndex());
 385
 386                 followerToSnapshot.markSendStatus(false);
 387             }
 388
 389             if (!wasLastChunk && followerToSnapshot.canSendNextChunk()) {
 390                 ActorSelection followerActor = context.getPeerActorSelection(followerId);
 391                 if(followerActor != null) {
 392                     sendSnapshotChunk(followerActor, followerId);
 393                 }
 394             }
 395
 396         } else {
 397             LOG.error("{}: Chunk index {} in InstallSnapshotReply from follower {} does not match expected index {}",
 398                     context.getId(), reply.getChunkIndex(), followerId,
 399                     followerToSnapshot.getChunkIndex());
 400
 401             if(reply.getChunkIndex() == INVALID_CHUNK_INDEX){
 402                 // Since the Follower did not find this index to be valid we should reset the follower snapshot
 403                 // so that Installing the snapshot can resume from the beginning
 404                 followerToSnapshot.reset();
 405             }
 406         }
 407     }
 408
 409     private void replicate(Replicate replicate) {
 410         long logIndex = replicate.getReplicatedLogEntry().getIndex();
 411
 412         if(LOG.isDebugEnabled()) {
 413             LOG.debug("{}: Replicate message {}", context.getId(), logIndex);
 414         }
 415
 416         // Create a tracker entry we will use this later to notify the
 417         // client actor
 418         trackerList.add(
 419             new ClientRequestTrackerImpl(replicate.getClientActor(),
 420                 replicate.getIdentifier(),
 421                 logIndex)
 422         );
 423
 424         if (followerToLog.isEmpty()) {
 425             context.setCommitIndex(logIndex);
 426             applyLogToStateMachine(logIndex);
 427         } else {
 428             sendAppendEntries();
 429         }
 430     }
 431
 432     private void sendAppendEntries() {
 433         // Send an AppendEntries to all followers
 434         long heartbeatInterval = context.getConfigParams().getHeartBeatInterval().toMillis();
 435         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 436             final String followerId = e.getKey();
 437             final FollowerLogInformation followerLogInformation = e.getValue();
 438             // This checks helps not to send a repeat message to the follower
 439             if(followerLogInformation.timeSinceLastActivity() >= heartbeatInterval) {
 440                 sendUpdatesToFollower(followerId, followerLogInformation, true);
 441             }
 442         }
 443     }
 444
 445     /**
 446      *
 447      * This method checks if any update needs to be sent to the given follower. This includes append log entries,
 448      * sending next snapshot chunk, and initiating a snapshot.
 449      * @return true if any update is sent, false otherwise
 450      */
 451
 452     private void sendUpdatesToFollower(String followerId, FollowerLogInformation followerLogInformation,
 453                                           boolean sendHeartbeat) {
 454
 455         ActorSelection followerActor = context.getPeerActorSelection(followerId);
 456         if (followerActor != null) {
 457             long followerNextIndex = followerLogInformation.getNextIndex();
 458             boolean isFollowerActive = followerLogInformation.isFollowerActive();
 459
 460             if (mapFollowerToSnapshot.get(followerId) != null) {
 461                 // if install snapshot is in process , then sent next chunk if possible
 462                 if (isFollowerActive && mapFollowerToSnapshot.get(followerId).canSendNextChunk()) {
 463                     sendSnapshotChunk(followerActor, followerId);
 464                 } else if(sendHeartbeat) {
 465                     // we send a heartbeat even if we have not received a reply for the last chunk
 466                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 467                         Collections.<ReplicatedLogEntry>emptyList(), followerId);
 468                 }
 469             } else {
 470                 long leaderLastIndex = context.getReplicatedLog().lastIndex();
 471                 long leaderSnapShotIndex = context.getReplicatedLog().getSnapshotIndex();
 472                 if (isFollowerActive &&
 473                     context.getReplicatedLog().isPresent(followerNextIndex)) {
 474                     // FIXME : Sending one entry at a time
 475                     final List<ReplicatedLogEntry> entries = context.getReplicatedLog().getFrom(followerNextIndex, 1);
 476
 477                     sendAppendEntriesToFollower(followerActor, followerNextIndex, entries, followerId);
 478
 479                 } else if (isFollowerActive && followerNextIndex >= 0 &&
 480                     leaderLastIndex >= followerNextIndex) {
 481                     // if the followers next index is not present in the leaders log, and
 482                     // if the follower is just not starting and if leader's index is more than followers index
 483                     // then snapshot should be sent
 484
 485                     if (LOG.isDebugEnabled()) {
 486                         LOG.debug("InitiateInstallSnapshot to follower:{}," +
 487                                 "follower-nextIndex:{}, leader-snapshot-index:{},  " +
 488                                 "leader-last-index:{}", followerId,
 489                             followerNextIndex, leaderSnapShotIndex, leaderLastIndex
 490                         );
 491                     }
 492                     actor().tell(new InitiateInstallSnapshot(), actor());
 493
 494                     // Send heartbeat to follower whenever install snapshot is initiated.
 495                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 496                             Collections.<ReplicatedLogEntry>emptyList(), followerId);
 497
 498                 } else if(sendHeartbeat) {
 499                     //we send an AppendEntries, even if the follower is inactive
 500                     // in-order to update the followers timestamp, in case it becomes active again
 501                     sendAppendEntriesToFollower(followerActor, followerLogInformation.getNextIndex(),
 502                         Collections.<ReplicatedLogEntry>emptyList(), followerId);
 503                 }
 504
 505             }
 506         }
 507     }
 508
 509     private void sendAppendEntriesToFollower(ActorSelection followerActor, long followerNextIndex,
 510         List<ReplicatedLogEntry> entries, String followerId) {
 511         AppendEntries appendEntries = new AppendEntries(currentTerm(), context.getId(),
 512             prevLogIndex(followerNextIndex),
 513             prevLogTerm(followerNextIndex), entries,
 514             context.getCommitIndex(), replicatedToAllIndex);
 515
 516         if(!entries.isEmpty()) {
 517             LOG.debug("{}: Sending AppendEntries to follower {}: {}", context.getId(), followerId,
 518                     appendEntries);
 519         }
 520
 521         followerActor.tell(appendEntries.toSerializable(), actor());
 522     }
 523
 524     /**
 525      * An installSnapshot is scheduled at a interval that is a multiple of
 526      * a HEARTBEAT_INTERVAL. This is to avoid the need to check for installing
 527      * snapshots at every heartbeat.
 528      *
 529      * Install Snapshot works as follows
 530      * 1. Leader sends a InitiateInstallSnapshot message to self
 531      * 2. Leader then initiates the capture snapshot by sending a CaptureSnapshot message to actor
 532      * 3. RaftActor on receipt of the CaptureSnapshotReply (from Shard), stores the received snapshot in the replicated log
 533      * and makes a call to Leader's handleMessage , with SendInstallSnapshot message.
 534      * 4. Leader , picks the snapshot from im-mem ReplicatedLog and sends it in chunks to the Follower
 535      * 5. On complete, Follower sends back a InstallSnapshotReply.
 536      * 6. On receipt of the InstallSnapshotReply for the last chunk, Leader marks the install complete for that follower
 537      * and replenishes the memory by deleting the snapshot in Replicated log.
 538      *
 539      */
 540     private void installSnapshotIfNeeded() {
 541         if(LOG.isDebugEnabled()) {
 542             LOG.debug("{}: installSnapshotIfNeeded, followers {}", context.getId(), followerToLog.keySet());
 543         }
 544
 545         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 546             final ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
 547
 548             if (followerActor != null) {
 549                 long nextIndex = e.getValue().getNextIndex();
 550
 551                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 552                         context.getReplicatedLog().isInSnapshot(nextIndex)) {
 553                     LOG.info("{}: {} follower needs a snapshot install", context.getId(), e.getKey());
 554                     if (snapshot.isPresent()) {
 555                         // if a snapshot is present in the memory, most likely another install is in progress
 556                         // no need to capture snapshot
 557                         sendSnapshotChunk(followerActor, e.getKey());
 558
 559                     } else if (!context.isSnapshotCaptureInitiated()) {
 560                         initiateCaptureSnapshot();
 561                         //we just need 1 follower who would need snapshot to be installed.
 562                         // when we have the snapshot captured, we would again check (in SendInstallSnapshot)
 563                         // who needs an install and send to all who need
 564                         break;
 565                     }
 566
 567                 }
 568             }
 569         }
 570     }
 571
 572     // on every install snapshot, we try to capture the snapshot.
 573     // Once a capture is going on, another one issued will get ignored by RaftActor.
 574     private void initiateCaptureSnapshot() {
 575         LOG.info("{}: Initiating Snapshot Capture to Install Snapshot, Leader:{}", context.getId(), getLeaderId());
 576         ReplicatedLogEntry lastAppliedEntry = context.getReplicatedLog().get(context.getLastApplied());
 577         long lastAppliedIndex = -1;
 578         long lastAppliedTerm = -1;
 579
 580         if (lastAppliedEntry != null) {
 581             lastAppliedIndex = lastAppliedEntry.getIndex();
 582             lastAppliedTerm = lastAppliedEntry.getTerm();
 583         } else if (context.getReplicatedLog().getSnapshotIndex() > -1)  {
 584             lastAppliedIndex = context.getReplicatedLog().getSnapshotIndex();
 585             lastAppliedTerm = context.getReplicatedLog().getSnapshotTerm();
 586         }
 587
 588         boolean isInstallSnapshotInitiated = true;
 589         actor().tell(new CaptureSnapshot(lastIndex(), lastTerm(),
 590                 lastAppliedIndex, lastAppliedTerm, isInstallSnapshotInitiated),
 591             actor());
 592         context.setSnapshotCaptureInitiated(true);
 593     }
 594
 595
 596     private void sendInstallSnapshot() {
 597         for (Entry<String, FollowerLogInformation> e : followerToLog.entrySet()) {
 598             ActorSelection followerActor = context.getPeerActorSelection(e.getKey());
 599
 600             if (followerActor != null) {
 601                 long nextIndex = e.getValue().getNextIndex();
 602
 603                 if (!context.getReplicatedLog().isPresent(nextIndex) &&
 604                     context.getReplicatedLog().isInSnapshot(nextIndex)) {
 605                     sendSnapshotChunk(followerActor, e.getKey());
 606                 }
 607             }
 608         }
 609     }
 610
 611     /**
 612      *  Sends a snapshot chunk to a given follower
 613      *  InstallSnapshot should qualify as a heartbeat too.
 614      */
 615     private void sendSnapshotChunk(ActorSelection followerActor, String followerId) {
 616         try {
 617             if (snapshot.isPresent()) {
 618                 ByteString nextSnapshotChunk = getNextSnapshotChunk(followerId,snapshot.get());
 619
 620                 // Note: the previous call to getNextSnapshotChunk has the side-effect of adding
 621                 // followerId to the followerToSnapshot map.
 622                 FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 623
 624                 followerActor.tell(
 625                     new InstallSnapshot(currentTerm(), context.getId(),
 626                         context.getReplicatedLog().getSnapshotIndex(),
 627                         context.getReplicatedLog().getSnapshotTerm(),
 628                         nextSnapshotChunk,
 629                             followerToSnapshot.incrementChunkIndex(),
 630                             followerToSnapshot.getTotalChunks(),
 631                         Optional.of(followerToSnapshot.getLastChunkHashCode())
 632                     ).toSerializable(),
 633                     actor()
 634                 );
 635                 LOG.info("{}: InstallSnapshot sent to follower {}, Chunk: {}/{}",
 636                         context.getId(), followerActor.path(),
 637                         followerToSnapshot.getChunkIndex(),
 638                         followerToSnapshot.getTotalChunks());
 639             }
 640         } catch (IOException e) {
 641             LOG.error(e, "{}: InstallSnapshot failed for Leader.", context.getId());
 642         }
 643     }
 644
 645     /**
 646      * Acccepts snaphot as ByteString, enters into map for future chunks
 647      * creates and return a ByteString chunk
 648      */
 649     private ByteString getNextSnapshotChunk(String followerId, ByteString snapshotBytes) throws IOException {
 650         FollowerToSnapshot followerToSnapshot = mapFollowerToSnapshot.get(followerId);
 651         if (followerToSnapshot == null) {
 652             followerToSnapshot = new FollowerToSnapshot(snapshotBytes);
 653             mapFollowerToSnapshot.put(followerId, followerToSnapshot);
 654         }
 655         ByteString nextChunk = followerToSnapshot.getNextChunk();
 656         if (LOG.isDebugEnabled()) {
 657             LOG.debug("{}: Leader's snapshot nextChunk size:{}", context.getId(), nextChunk.size());
 658         }
 659         return nextChunk;
 660     }
 661
 662     private void sendHeartBeat() {
 663         if (!followerToLog.isEmpty()) {
 664             sendAppendEntries();
 665         }
 666     }
 667
 668     private void stopHeartBeat() {
 669         if (heartbeatSchedule != null && !heartbeatSchedule.isCancelled()) {
 670             heartbeatSchedule.cancel();
 671         }
 672     }
 673
 674     private void scheduleHeartBeat(FiniteDuration interval) {
 675         if (followerToLog.isEmpty()) {
 676             // Optimization - do not bother scheduling a heartbeat as there are
 677             // no followers
 678             return;
 679         }
 680
 681         stopHeartBeat();
 682
 683         // Schedule a heartbeat. When the scheduler triggers a SendHeartbeat
 684         // message is sent to itself.
 685         // Scheduling the heartbeat only once here because heartbeats do not
 686         // need to be sent if there are other messages being sent to the remote
 687         // actor.
 688         heartbeatSchedule = context.getActorSystem().scheduler().scheduleOnce(
 689             interval, context.getActor(), new SendHeartBeat(),
 690             context.getActorSystem().dispatcher(), context.getActor());
 691     }
 692
 693     @Override
 694     public void close() throws Exception {
 695         stopHeartBeat();
 696     }
 697
 698     @Override
 699     public String getLeaderId() {
 700         return context.getId();
 701     }
 702
 703     protected boolean isLeaderIsolated() {
 704         int minPresent = minIsolatedLeaderPeerCount;
 705         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 706             if (followerLogInformation.isFollowerActive()) {
 707                 --minPresent;
 708                 if (minPresent == 0) {
 709                     break;
 710                 }
 711             }
 712         }
 713         return (minPresent != 0);
 714     }
 715
 716     /**
 717      * Encapsulates the snapshot bytestring and handles the logic of sending
 718      * snapshot chunks
 719      */
 720     protected class FollowerToSnapshot {
 721         private final ByteString snapshotBytes;
 722         private int offset = 0;
 723         // the next snapshot chunk is sent only if the replyReceivedForOffset matches offset
 724         private int replyReceivedForOffset;
 725         // if replyStatus is false, the previous chunk is attempted
 726         private boolean replyStatus = false;
 727         private int chunkIndex;
 728         private final int totalChunks;
 729         private int lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 730         private int nextChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 731
 732         public FollowerToSnapshot(ByteString snapshotBytes) {
 733             this.snapshotBytes = snapshotBytes;
 734             int size = snapshotBytes.size();
 735             totalChunks = ( size / context.getConfigParams().getSnapshotChunkSize()) +
 736                 ((size % context.getConfigParams().getSnapshotChunkSize()) > 0 ? 1 : 0);
 737             if(LOG.isDebugEnabled()) {
 738                 LOG.debug("{}: Snapshot {} bytes, total chunks to send:{}",
 739                         context.getId(), size, totalChunks);
 740             }
 741             replyReceivedForOffset = -1;
 742             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 743         }
 744
 745         public ByteString getSnapshotBytes() {
 746             return snapshotBytes;
 747         }
 748
 749         public int incrementOffset() {
 750             if(replyStatus) {
 751                 // if prev chunk failed, we would want to sent the same chunk again
 752                 offset = offset + context.getConfigParams().getSnapshotChunkSize();
 753             }
 754             return offset;
 755         }
 756
 757         public int incrementChunkIndex() {
 758             if (replyStatus) {
 759                 // if prev chunk failed, we would want to sent the same chunk again
 760                 chunkIndex =  chunkIndex + 1;
 761             }
 762             return chunkIndex;
 763         }
 764
 765         public int getChunkIndex() {
 766             return chunkIndex;
 767         }
 768
 769         public int getTotalChunks() {
 770             return totalChunks;
 771         }
 772
 773         public boolean canSendNextChunk() {
 774             // we only send a false if a chunk is sent but we have not received a reply yet
 775             return replyReceivedForOffset == offset;
 776         }
 777
 778         public boolean isLastChunk(int chunkIndex) {
 779             return totalChunks == chunkIndex;
 780         }
 781
 782         public void markSendStatus(boolean success) {
 783             if (success) {
 784                 // if the chunk sent was successful
 785                 replyReceivedForOffset = offset;
 786                 replyStatus = true;
 787                 lastChunkHashCode = nextChunkHashCode;
 788             } else {
 789                 // if the chunk sent was failure
 790                 replyReceivedForOffset = offset;
 791                 replyStatus = false;
 792             }
 793         }
 794
 795         public ByteString getNextChunk() {
 796             int snapshotLength = getSnapshotBytes().size();
 797             int start = incrementOffset();
 798             int size = context.getConfigParams().getSnapshotChunkSize();
 799             if (context.getConfigParams().getSnapshotChunkSize() > snapshotLength) {
 800                 size = snapshotLength;
 801             } else {
 802                 if ((start + context.getConfigParams().getSnapshotChunkSize()) > snapshotLength) {
 803                     size = snapshotLength - start;
 804                 }
 805             }
 806
 807             if(LOG.isDebugEnabled()) {
 808                 LOG.debug("{}: Next chunk: length={}, offset={},size={}", context.getId(),
 809                     snapshotLength, start, size);
 810             }
 811             ByteString substring = getSnapshotBytes().substring(start, start + size);
 812             nextChunkHashCode = substring.hashCode();
 813             return substring;
 814         }
 815
 816         /**
 817          * reset should be called when the Follower needs to be sent the snapshot from the beginning
 818          */
 819         public void reset(){
 820             offset = 0;
 821             replyStatus = false;
 822             replyReceivedForOffset = offset;
 823             chunkIndex = AbstractLeader.FIRST_CHUNK_INDEX;
 824             lastChunkHashCode = AbstractLeader.INITIAL_LAST_CHUNK_HASH_CODE;
 825         }
 826
 827         public int getLastChunkHashCode() {
 828             return lastChunkHashCode;
 829         }
 830     }
 831
 832     // called from example-actor for printing the follower-states
 833     public String printFollowerStates() {
 834         final StringBuilder sb = new StringBuilder();
 835
 836         sb.append('[');
 837         for (FollowerLogInformation followerLogInformation : followerToLog.values()) {
 838             sb.append('{');
 839             sb.append(followerLogInformation.getId());
 840             sb.append(" state:");
 841             sb.append(followerLogInformation.isFollowerActive());
 842             sb.append("},");
 843         }
 844         sb.append(']');
 845
 846         return sb.toString();
 847     }
 848
 849     @VisibleForTesting
 850     public FollowerLogInformation getFollower(String followerId) {
 851         return followerToLog.get(followerId);
 852     }
 853
 854     @VisibleForTesting
 855     protected void setFollowerSnapshot(String followerId, FollowerToSnapshot snapshot) {
 856         mapFollowerToSnapshot.put(followerId, snapshot);
 857     }
 858
 859     @VisibleForTesting
 860     public int followerSnapshotSize() {
 861         return mapFollowerToSnapshot.size();
 862     }
 863
 864     @VisibleForTesting
 865     public int followerLogSize() {
 866         return followerToLog.size();
 867     }
 868 }